mirror of
https://github.com/Niko1221/Strata.git
synced 2026-11-09 03:59:45 -10:00
tests/snapshot: golden files of main 146df3c46 (identical when made on Windows and Linux)
This commit is contained in:
Niko1221 committed
1 parent
affa7d96e7
commit
01f9c62a0a
32 files changed
+10249
No files matched your search
@@ -0,0 +1,257 @@
|
||||
{
|
||||
"machine": "arc_a750",
|
||||
"set": "deepseek",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family deepseek41 --model IST-DAS-2.00bpw --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model IST-DAS-2.00bpw --yes --no-tui --context 131072",
|
||||
"fetches": [],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"deepseek41",
|
||||
"--model",
|
||||
"IST-DAS-2.00bpw",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"yes",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/<STATE>/launch/ist-das-2-00bpw-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"engine_ready_s": 3600,
|
||||
"backend": "sycl",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/<STATE>/launch/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load deepseek-v4.1-flash-ist-das-2.00bpw",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": null
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/install/strata-deepseek-v4.1-flash-ist-das-2.00bpw.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"engine_ready_s": 3600,
|
||||
"backend": "sycl",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"aliases": [
|
||||
"deepseek-v4.1-flash-ist-das-2.00bpw"
|
||||
],
|
||||
"api_key": "<KEY>"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/install/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load deepseek-v4.1-flash-ist-das-2.00bpw with Drafting = Auto",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": null
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/install/strata-deepseek-v4.1-flash-ist-das-2.00bpw.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"engine_ready_s": 3600,
|
||||
"backend": "sycl",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"aliases": [
|
||||
"deepseek-v4.1-flash-ist-das-2.00bpw"
|
||||
],
|
||||
"api_key": "<KEY>"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/install/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,343 @@
|
||||
{
|
||||
"machine": "arc_a750",
|
||||
"set": "iq3_s",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family qwen --model IQ3_S --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model IQ3_S --yes --no-tui --context 131072",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure",
|
||||
"file": "qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"gb": 1.2
|
||||
}
|
||||
],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"qwen",
|
||||
"--model",
|
||||
"IQ3_S",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"cpu",
|
||||
"--experimental-speed-projection",
|
||||
"off",
|
||||
"--drafter",
|
||||
"dflash2",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/<STATE>/launch/iq3_s-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"backend": "sycl",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1008,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.6"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.6)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load iq3_s",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 1.2,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/install/strata-iq3_s.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "sycl",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"iq3_s"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] Drafting: MTP (the DFlash2 drafter isn't downloaded yet)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load iq3_s with Drafting = Auto",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure",
|
||||
"file": "qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"gb": 1.2
|
||||
}
|
||||
],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 1.2,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "draftingFetch",
|
||||
"mode": "auto",
|
||||
"noSplit": false
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/install/strata-iq3_s.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "sycl",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"iq3_s"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.6"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.6)"
|
||||
]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,341 @@
|
||||
{
|
||||
"machine": "arc_a750",
|
||||
"set": "ud_iq4_xs_v31",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family unsloth --model UD-IQ4_XS-v31 --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model UD-IQ4_XS-v31 --yes --no-tui --context 131072",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure",
|
||||
"file": "qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"gb": 1.2
|
||||
}
|
||||
],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"unsloth",
|
||||
"--model",
|
||||
"UD-IQ4_XS-v31",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"cpu",
|
||||
"--drafter",
|
||||
"dflash2",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/<STATE>/launch/ud-iq4_xs-v31-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"backend": "sycl",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1089,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.6"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.6)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load unsloth-ud-iq4_xs-v31",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 1.2,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-iq4_xs-v31.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "sycl",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-iq4_xs-v31"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] Drafting: MTP (the DFlash2 drafter isn't downloaded yet)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load unsloth-ud-iq4_xs-v31 with Drafting = Auto",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure",
|
||||
"file": "qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"gb": 1.2
|
||||
}
|
||||
],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 1.2,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "draftingFetch",
|
||||
"mode": "auto",
|
||||
"noSplit": false
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-iq4_xs-v31.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "sycl",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-iq4_xs-v31"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.6"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.6)"
|
||||
]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,341 @@
|
||||
{
|
||||
"machine": "arc_a750",
|
||||
"set": "ud_q4_k_xl",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family unsloth --model UD-Q4_K_XL --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model UD-Q4_K_XL --yes --no-tui --context 131072",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure",
|
||||
"file": "qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"gb": 1.2
|
||||
}
|
||||
],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"unsloth",
|
||||
"--model",
|
||||
"UD-Q4_K_XL",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"cpu",
|
||||
"--drafter",
|
||||
"dflash2",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/<STATE>/launch/ud-q4_k_xl-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"backend": "sycl",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1230,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.6"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.6)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load unsloth-ud-q4_k_xl",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 1.2,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-q4_k_xl.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "sycl",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-q4_k_xl"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] Drafting: MTP (the DFlash2 drafter isn't downloaded yet)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load unsloth-ud-q4_k_xl with Drafting = Auto",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure",
|
||||
"file": "qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"gb": 1.2
|
||||
}
|
||||
],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 1.2,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "draftingFetch",
|
||||
"mode": "auto",
|
||||
"noSplit": false
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-q4_k_xl.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "sycl",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-q4_k_xl"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.6"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.6)"
|
||||
]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,254 @@
|
||||
{
|
||||
"machine": "dgx_spark",
|
||||
"set": "deepseek",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family deepseek41 --model IST-DAS-2.00bpw --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model IST-DAS-2.00bpw --yes --no-tui --context 131072",
|
||||
"fetches": [],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"deepseek41",
|
||||
"--model",
|
||||
"IST-DAS-2.00bpw",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"yes",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/<STATE>/launch/ist-das-2-00bpw-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"engine_ready_s": 3600,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/<STATE>/launch/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load deepseek-v4.1-flash-ist-das-2.00bpw",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": null
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/install/strata-deepseek-v4.1-flash-ist-das-2.00bpw.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"engine_ready_s": 3600,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"aliases": [
|
||||
"deepseek-v4.1-flash-ist-das-2.00bpw"
|
||||
],
|
||||
"api_key": "<KEY>"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/install/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load deepseek-v4.1-flash-ist-das-2.00bpw with Drafting = Auto",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": null
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/install/strata-deepseek-v4.1-flash-ist-das-2.00bpw.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"engine_ready_s": 3600,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"aliases": [
|
||||
"deepseek-v4.1-flash-ist-das-2.00bpw"
|
||||
],
|
||||
"api_key": "<KEY>"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/install/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,375 @@
|
||||
{
|
||||
"machine": "dgx_spark",
|
||||
"set": "iq3_s",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family qwen --model IQ3_S --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model IQ3_S --yes --no-tui --context 131072",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "cuda",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"qwen",
|
||||
"--model",
|
||||
"IQ3_S",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"gpu",
|
||||
"--experimental-speed-projection",
|
||||
"off",
|
||||
"--drafter",
|
||||
"dflash2",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto:32768",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--mmap-experts",
|
||||
"--mtp-q4",
|
||||
"all",
|
||||
"--mtp-hnorm",
|
||||
"stream"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/<STATE>/launch/iq3_s-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1008,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto:32768",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--mmap-experts",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.35"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.35)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load iq3_s",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto:32768",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--mmap-experts",
|
||||
"--mtp-q4",
|
||||
"all",
|
||||
"--mtp-hnorm",
|
||||
"stream"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/install/strata-iq3_s.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"iq3_s"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto:32768",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--mmap-experts",
|
||||
"--mtp-q4",
|
||||
"all",
|
||||
"--mtp-hnorm",
|
||||
"stream"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] Drafting: MTP (the DFlash2 drafter isn't downloaded yet)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load iq3_s with Drafting = Auto",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "cuda",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "draftingFetch",
|
||||
"mode": "auto",
|
||||
"noSplit": false
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto:32768",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--mmap-experts",
|
||||
"--mtp-q4",
|
||||
"all",
|
||||
"--mtp-hnorm",
|
||||
"stream"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/install/strata-iq3_s.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"iq3_s"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto:32768",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--mmap-experts",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.35"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.35)"
|
||||
]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,373 @@
|
||||
{
|
||||
"machine": "dgx_spark",
|
||||
"set": "ud_iq4_xs_v31",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family unsloth --model UD-IQ4_XS-v31 --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model UD-IQ4_XS-v31 --yes --no-tui --context 131072",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "cuda",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"unsloth",
|
||||
"--model",
|
||||
"UD-IQ4_XS-v31",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"gpu",
|
||||
"--drafter",
|
||||
"dflash2",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto:32768",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--mmap-experts",
|
||||
"--mtp-q4",
|
||||
"all",
|
||||
"--mtp-hnorm",
|
||||
"stream",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/<STATE>/launch/ud-iq4_xs-v31-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1089,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto:32768",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--mmap-experts",
|
||||
"--resident-budget-gib",
|
||||
"48",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.35"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.35)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load unsloth-ud-iq4_xs-v31",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto:32768",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--mmap-experts",
|
||||
"--mtp-q4",
|
||||
"all",
|
||||
"--mtp-hnorm",
|
||||
"stream",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-iq4_xs-v31.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-iq4_xs-v31"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto:32768",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--mmap-experts",
|
||||
"--mtp-q4",
|
||||
"all",
|
||||
"--mtp-hnorm",
|
||||
"stream",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] Drafting: MTP (the DFlash2 drafter isn't downloaded yet)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load unsloth-ud-iq4_xs-v31 with Drafting = Auto",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "cuda",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "draftingFetch",
|
||||
"mode": "auto",
|
||||
"noSplit": false
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto:32768",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--mmap-experts",
|
||||
"--mtp-q4",
|
||||
"all",
|
||||
"--mtp-hnorm",
|
||||
"stream",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-iq4_xs-v31.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-iq4_xs-v31"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto:32768",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--mmap-experts",
|
||||
"--resident-budget-gib",
|
||||
"48",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.35"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.35)"
|
||||
]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,369 @@
|
||||
{
|
||||
"machine": "dgx_spark",
|
||||
"set": "ud_q4_k_xl",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family unsloth --model UD-Q4_K_XL --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model UD-Q4_K_XL --yes --no-tui --context 131072",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure",
|
||||
"file": "qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"gb": 1.2
|
||||
}
|
||||
],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"unsloth",
|
||||
"--model",
|
||||
"UD-Q4_K_XL",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"gpu",
|
||||
"--drafter",
|
||||
"dflash2",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto:32768",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--mmap-experts",
|
||||
"--mtp-q4",
|
||||
"all",
|
||||
"--mtp-hnorm",
|
||||
"stream",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/<STATE>/launch/ud-q4_k_xl-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1230,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto:32768",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--mmap-experts",
|
||||
"--resident-budget-gib",
|
||||
"48",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.35"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.35)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load unsloth-ud-q4_k_xl",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 1.2,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto:32768",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--mmap-experts",
|
||||
"--mtp-q4",
|
||||
"all",
|
||||
"--mtp-hnorm",
|
||||
"stream",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-q4_k_xl.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-q4_k_xl"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto:32768",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--mmap-experts",
|
||||
"--mtp-q4",
|
||||
"all",
|
||||
"--mtp-hnorm",
|
||||
"stream",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] Drafting: MTP (the DFlash2 drafter isn't downloaded yet)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load unsloth-ud-q4_k_xl with Drafting = Auto",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure",
|
||||
"file": "qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"gb": 1.2
|
||||
}
|
||||
],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 1.2,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "draftingFetch",
|
||||
"mode": "auto",
|
||||
"noSplit": false
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto:32768",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--mmap-experts",
|
||||
"--mtp-q4",
|
||||
"all",
|
||||
"--mtp-hnorm",
|
||||
"stream",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-q4_k_xl.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-q4_k_xl"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto:32768",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--mmap-experts",
|
||||
"--resident-budget-gib",
|
||||
"48",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.35"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.35)"
|
||||
]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,245 @@
|
||||
{
|
||||
"machine": "mac_large",
|
||||
"set": "deepseek",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family deepseek41 --model IST-DAS-2.00bpw --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model IST-DAS-2.00bpw --yes --no-tui --context 131072",
|
||||
"fetches": [],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"deepseek41",
|
||||
"--model",
|
||||
"IST-DAS-2.00bpw",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"yes",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"none"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/<STATE>/launch/ist-das-2-00bpw-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"engine_ready_s": 3600,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"none"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/<STATE>/launch/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load deepseek-v4.1-flash-ist-das-2.00bpw",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": null
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"none"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/install/strata-deepseek-v4.1-flash-ist-das-2.00bpw.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"engine_ready_s": 3600,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"aliases": [
|
||||
"deepseek-v4.1-flash-ist-das-2.00bpw"
|
||||
],
|
||||
"api_key": "<KEY>"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"none"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/install/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load deepseek-v4.1-flash-ist-das-2.00bpw with Drafting = Auto",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": null
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"none"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/install/strata-deepseek-v4.1-flash-ist-das-2.00bpw.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"engine_ready_s": 3600,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"aliases": [
|
||||
"deepseek-v4.1-flash-ist-das-2.00bpw"
|
||||
],
|
||||
"api_key": "<KEY>"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"none"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/install/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,304 @@
|
||||
{
|
||||
"machine": "mac_large",
|
||||
"set": "iq3_s",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family qwen --model IQ3_S --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model IQ3_S --yes --no-tui --context 131072",
|
||||
"fetches": [],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"qwen",
|
||||
"--model",
|
||||
"IQ3_S",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"gpu",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/<STATE>/launch/iq3_s-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1008,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load iq3_s",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/install/strata-iq3_s.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"iq3_s"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load iq3_s with Drafting = Auto",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/install/strata-iq3_s.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"iq3_s"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": []
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,343 @@
|
||||
{
|
||||
"machine": "mac_large",
|
||||
"set": "ud_iq4_xs_v31",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family unsloth --model UD-IQ4_XS-v31 --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model UD-IQ4_XS-v31 --yes --no-tui --context 131072",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "metal",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"unsloth",
|
||||
"--model",
|
||||
"UD-IQ4_XS-v31",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"gpu",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/<STATE>/launch/ud-iq4_xs-v31-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1089,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.7"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.7)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load unsloth-ud-iq4_xs-v31",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-iq4_xs-v31.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-iq4_xs-v31"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] Drafting: MTP (the DFlash2 drafter isn't downloaded yet)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load unsloth-ud-iq4_xs-v31 with Drafting = Auto",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "metal",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "draftingFetch",
|
||||
"mode": "auto",
|
||||
"noSplit": false
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-iq4_xs-v31.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-iq4_xs-v31"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.7"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.7)"
|
||||
]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,343 @@
|
||||
{
|
||||
"machine": "mac_large",
|
||||
"set": "ud_q4_k_xl",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family unsloth --model UD-Q4_K_XL --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model UD-Q4_K_XL --yes --no-tui --context 131072",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "metal",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"unsloth",
|
||||
"--model",
|
||||
"UD-Q4_K_XL",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"gpu",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/<STATE>/launch/ud-q4_k_xl-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1230,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.7"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.7)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load unsloth-ud-q4_k_xl",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-q4_k_xl.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-q4_k_xl"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] Drafting: MTP (the DFlash2 drafter isn't downloaded yet)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load unsloth-ud-q4_k_xl with Drafting = Auto",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "metal",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "draftingFetch",
|
||||
"mode": "auto",
|
||||
"noSplit": false
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-q4_k_xl.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-q4_k_xl"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.7"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.7)"
|
||||
]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,245 @@
|
||||
{
|
||||
"machine": "mac_small",
|
||||
"set": "deepseek",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family deepseek41 --model IST-DAS-2.00bpw --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model IST-DAS-2.00bpw --yes --no-tui --context 131072",
|
||||
"fetches": [],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"deepseek41",
|
||||
"--model",
|
||||
"IST-DAS-2.00bpw",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"yes",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"none"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/<STATE>/launch/ist-das-2-00bpw-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"engine_ready_s": 3600,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"none"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/<STATE>/launch/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load deepseek-v4.1-flash-ist-das-2.00bpw",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": null
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"none"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/install/strata-deepseek-v4.1-flash-ist-das-2.00bpw.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"engine_ready_s": 3600,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"aliases": [
|
||||
"deepseek-v4.1-flash-ist-das-2.00bpw"
|
||||
],
|
||||
"api_key": "<KEY>"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"none"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/install/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load deepseek-v4.1-flash-ist-das-2.00bpw with Drafting = Auto",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": null
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"none"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/install/strata-deepseek-v4.1-flash-ist-das-2.00bpw.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"engine_ready_s": 3600,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"aliases": [
|
||||
"deepseek-v4.1-flash-ist-das-2.00bpw"
|
||||
],
|
||||
"api_key": "<KEY>"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"none"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/install/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,304 @@
|
||||
{
|
||||
"machine": "mac_small",
|
||||
"set": "iq3_s",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family qwen --model IQ3_S --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model IQ3_S --yes --no-tui --context 32768",
|
||||
"fetches": [],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"qwen",
|
||||
"--model",
|
||||
"IQ3_S",
|
||||
"--context",
|
||||
"32768",
|
||||
"--vision",
|
||||
"gpu",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"32768",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/<STATE>/launch/iq3_s-32k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1008,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"32768",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load iq3_s",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"32768",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/install/strata-iq3_s.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"iq3_s"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"32768",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load iq3_s with Drafting = Auto",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"32768",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/install/strata-iq3_s.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"iq3_s"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"32768",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": []
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,343 @@
|
||||
{
|
||||
"machine": "mac_small",
|
||||
"set": "ud_iq4_xs_v31",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family unsloth --model UD-IQ4_XS-v31 --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model UD-IQ4_XS-v31 --yes --no-tui --context 32768",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "metal",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"unsloth",
|
||||
"--model",
|
||||
"UD-IQ4_XS-v31",
|
||||
"--context",
|
||||
"32768",
|
||||
"--vision",
|
||||
"gpu",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"32768",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/<STATE>/launch/ud-iq4_xs-v31-32k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1089,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"32768",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.7"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.7)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load unsloth-ud-iq4_xs-v31",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"32768",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-iq4_xs-v31.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-iq4_xs-v31"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"32768",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] Drafting: MTP (the DFlash2 drafter isn't downloaded yet)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load unsloth-ud-iq4_xs-v31 with Drafting = Auto",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "metal",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "draftingFetch",
|
||||
"mode": "auto",
|
||||
"noSplit": false
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"32768",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-iq4_xs-v31.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-iq4_xs-v31"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"32768",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.7"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.7)"
|
||||
]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,343 @@
|
||||
{
|
||||
"machine": "mac_small",
|
||||
"set": "ud_q4_k_xl",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family unsloth --model UD-Q4_K_XL --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model UD-Q4_K_XL --yes --no-tui --context 32768",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "metal",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"unsloth",
|
||||
"--model",
|
||||
"UD-Q4_K_XL",
|
||||
"--context",
|
||||
"32768",
|
||||
"--vision",
|
||||
"gpu",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"32768",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/<STATE>/launch/ud-q4_k_xl-32k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1230,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"32768",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.7"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.7)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load unsloth-ud-q4_k_xl",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"32768",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-q4_k_xl.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-q4_k_xl"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"32768",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] Drafting: MTP (the DFlash2 drafter isn't downloaded yet)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load unsloth-ud-q4_k_xl with Drafting = Auto",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "metal",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "draftingFetch",
|
||||
"mode": "auto",
|
||||
"noSplit": false
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"32768",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-q4_k_xl.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "metal",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-q4_k_xl"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"32768",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"24",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.7"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.7)"
|
||||
]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,254 @@
|
||||
{
|
||||
"machine": "nv_12gb_streaming",
|
||||
"set": "deepseek",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family deepseek41 --model IST-DAS-2.00bpw --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model IST-DAS-2.00bpw --yes --no-tui --context 131072",
|
||||
"fetches": [],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"deepseek41",
|
||||
"--model",
|
||||
"IST-DAS-2.00bpw",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"yes",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/<STATE>/launch/ist-das-2-00bpw-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"engine_ready_s": 3600,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/<STATE>/launch/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load deepseek-v4.1-flash-ist-das-2.00bpw",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": null
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/install/strata-deepseek-v4.1-flash-ist-das-2.00bpw.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"engine_ready_s": 3600,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"aliases": [
|
||||
"deepseek-v4.1-flash-ist-das-2.00bpw"
|
||||
],
|
||||
"api_key": "<KEY>"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/install/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load deepseek-v4.1-flash-ist-das-2.00bpw with Drafting = Auto",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": null
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/install/strata-deepseek-v4.1-flash-ist-das-2.00bpw.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"engine_ready_s": 3600,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"aliases": [
|
||||
"deepseek-v4.1-flash-ist-das-2.00bpw"
|
||||
],
|
||||
"api_key": "<KEY>"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/install/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,326 @@
|
||||
{
|
||||
"machine": "nv_12gb_streaming",
|
||||
"set": "iq3_s",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family qwen --model IQ3_S --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model IQ3_S --yes --no-tui --context 65536",
|
||||
"fetches": [],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"qwen",
|
||||
"--model",
|
||||
"IQ3_S",
|
||||
"--context",
|
||||
"65536",
|
||||
"--vision",
|
||||
"cpu",
|
||||
"--experimental-speed-projection",
|
||||
"off",
|
||||
"--drafter",
|
||||
"dflash2",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"65536",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--kv-resident",
|
||||
"32768"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/<STATE>/launch/iq3_s-64k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1008,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"65536",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--kv-resident",
|
||||
"32768"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load iq3_s",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"65536",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--kv-resident",
|
||||
"32768"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/install/strata-iq3_s.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"iq3_s"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"65536",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--kv-resident",
|
||||
"32768"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load iq3_s with Drafting = Auto",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"65536",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--kv-resident",
|
||||
"32768"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/install/strata-iq3_s.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"iq3_s"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"65536",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--kv-resident",
|
||||
"32768"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,363 @@
|
||||
{
|
||||
"machine": "nv_12gb_streaming",
|
||||
"set": "ud_iq4_xs_v31",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family unsloth --model UD-IQ4_XS-v31 --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model UD-IQ4_XS-v31 --yes --no-tui --context 65536",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "cuda",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"unsloth",
|
||||
"--model",
|
||||
"UD-IQ4_XS-v31",
|
||||
"--context",
|
||||
"65536",
|
||||
"--vision",
|
||||
"cpu",
|
||||
"--drafter",
|
||||
"dflash2",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"65536",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--kv-resident",
|
||||
"32768",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/<STATE>/launch/ud-iq4_xs-v31-64k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1089,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"65536",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--kv-resident",
|
||||
"32768",
|
||||
"--resident-budget-gib",
|
||||
"24",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.8"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.8)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load unsloth-ud-iq4_xs-v31",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"65536",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--kv-resident",
|
||||
"32768",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-iq4_xs-v31.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-iq4_xs-v31"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"65536",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--kv-resident",
|
||||
"32768",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] Drafting: MTP (the DFlash2 drafter isn't downloaded yet)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load unsloth-ud-iq4_xs-v31 with Drafting = Auto",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "cuda",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "draftingFetch",
|
||||
"mode": "auto",
|
||||
"noSplit": false
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"65536",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--kv-resident",
|
||||
"32768",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-iq4_xs-v31.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-iq4_xs-v31"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"65536",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--kv-resident",
|
||||
"32768",
|
||||
"--resident-budget-gib",
|
||||
"24",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.8"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.8)"
|
||||
]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,359 @@
|
||||
{
|
||||
"machine": "nv_12gb_streaming",
|
||||
"set": "ud_q4_k_xl",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family unsloth --model UD-Q4_K_XL --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model UD-Q4_K_XL --yes --no-tui --context 65536",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure",
|
||||
"file": "qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"gb": 1.2
|
||||
}
|
||||
],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"unsloth",
|
||||
"--model",
|
||||
"UD-Q4_K_XL",
|
||||
"--context",
|
||||
"65536",
|
||||
"--vision",
|
||||
"cpu",
|
||||
"--drafter",
|
||||
"dflash2",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"65536",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--kv-resident",
|
||||
"32768",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/<STATE>/launch/ud-q4_k_xl-64k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1230,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"65536",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--kv-resident",
|
||||
"32768",
|
||||
"--resident-budget-gib",
|
||||
"24",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.8"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.8)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load unsloth-ud-q4_k_xl",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 1.2,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"65536",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--kv-resident",
|
||||
"32768",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-q4_k_xl.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-q4_k_xl"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"65536",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--kv-resident",
|
||||
"32768",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] Drafting: MTP (the DFlash2 drafter isn't downloaded yet)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load unsloth-ud-q4_k_xl with Drafting = Auto",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure",
|
||||
"file": "qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"gb": 1.2
|
||||
}
|
||||
],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 1.2,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "draftingFetch",
|
||||
"mode": "auto",
|
||||
"noSplit": false
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"65536",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--kv-resident",
|
||||
"32768",
|
||||
"--resident-budget-gib",
|
||||
"24"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-q4_k_xl.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-q4_k_xl"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"65536",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--kv-resident",
|
||||
"32768",
|
||||
"--resident-budget-gib",
|
||||
"24",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.8"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.8)"
|
||||
]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,254 @@
|
||||
{
|
||||
"machine": "nv_resident",
|
||||
"set": "deepseek",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family deepseek41 --model IST-DAS-2.00bpw --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model IST-DAS-2.00bpw --yes --no-tui --context 131072",
|
||||
"fetches": [],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"deepseek41",
|
||||
"--model",
|
||||
"IST-DAS-2.00bpw",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"yes",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/<STATE>/launch/ist-das-2-00bpw-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"engine_ready_s": 3600,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/<STATE>/launch/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load deepseek-v4.1-flash-ist-das-2.00bpw",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": null
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/install/strata-deepseek-v4.1-flash-ist-das-2.00bpw.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"engine_ready_s": 3600,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"aliases": [
|
||||
"deepseek-v4.1-flash-ist-das-2.00bpw"
|
||||
],
|
||||
"api_key": "<KEY>"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/install/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load deepseek-v4.1-flash-ist-das-2.00bpw with Drafting = Auto",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": null
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/install/strata-deepseek-v4.1-flash-ist-das-2.00bpw.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"engine_ready_s": 3600,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"aliases": [
|
||||
"deepseek-v4.1-flash-ist-das-2.00bpw"
|
||||
],
|
||||
"api_key": "<KEY>"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/install/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,367 @@
|
||||
{
|
||||
"machine": "nv_resident",
|
||||
"set": "iq3_s",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family qwen --model IQ3_S --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model IQ3_S --yes --no-tui --context 131072",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "cuda",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"qwen",
|
||||
"--model",
|
||||
"IQ3_S",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"gpu",
|
||||
"--experimental-speed-projection",
|
||||
"off",
|
||||
"--drafter",
|
||||
"dflash2",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/<STATE>/launch/iq3_s-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1008,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.35"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.35)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load iq3_s",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--batch-mtp"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/install/strata-iq3_s.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"parallel": 3,
|
||||
"aliases": [
|
||||
"iq3_s"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--batch-mtp",
|
||||
"--batch",
|
||||
"3"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_BATCH_ATTN_ROWS": "1",
|
||||
"STRATA_BATCH_GDN_GROUPS": "1",
|
||||
"STRATA_BATCH_SELF_COMMIT": "1",
|
||||
"STRATA_CKPT_POOL": "1",
|
||||
"STRATA_CKPT_POOL_READY": "4",
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1",
|
||||
"STRATA_QSA_SLOTS": "1",
|
||||
"STRATA_SLOT_CKPT_MOVE": "1",
|
||||
"STRATA_SLOT_COPY_BATCHSYNC": "1",
|
||||
"STRATA_SLOT_COPY_DIRECT": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] Drafting: MTP (DFlash2 isn't available in this mode)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load iq3_s with Drafting = Auto",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "cuda",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "draftingFetch",
|
||||
"mode": "auto",
|
||||
"noSplit": false
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/install/strata-iq3_s.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"iq3_s"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.35"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.35)"
|
||||
]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,365 @@
|
||||
{
|
||||
"machine": "nv_resident",
|
||||
"set": "ud_iq4_xs_v31",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family unsloth --model UD-IQ4_XS-v31 --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model UD-IQ4_XS-v31 --yes --no-tui --context 131072",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "cuda",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"unsloth",
|
||||
"--model",
|
||||
"UD-IQ4_XS-v31",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"gpu",
|
||||
"--drafter",
|
||||
"dflash2",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/<STATE>/launch/ud-iq4_xs-v31-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1089,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.35"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.35)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load unsloth-ud-iq4_xs-v31",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48",
|
||||
"--batch-mtp"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-iq4_xs-v31.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"parallel": 3,
|
||||
"aliases": [
|
||||
"unsloth-ud-iq4_xs-v31"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48",
|
||||
"--batch-mtp",
|
||||
"--batch",
|
||||
"3"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_BATCH_ATTN_ROWS": "1",
|
||||
"STRATA_BATCH_GDN_GROUPS": "1",
|
||||
"STRATA_BATCH_SELF_COMMIT": "1",
|
||||
"STRATA_CKPT_POOL": "1",
|
||||
"STRATA_CKPT_POOL_READY": "4",
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1",
|
||||
"STRATA_QSA_SLOTS": "1",
|
||||
"STRATA_SLOT_CKPT_MOVE": "1",
|
||||
"STRATA_SLOT_COPY_BATCHSYNC": "1",
|
||||
"STRATA_SLOT_COPY_DIRECT": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] Drafting: MTP (DFlash2 isn't available in this mode)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load unsloth-ud-iq4_xs-v31 with Drafting = Auto",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "cuda",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "draftingFetch",
|
||||
"mode": "auto",
|
||||
"noSplit": false
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-iq4_xs-v31.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-iq4_xs-v31"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.35"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.35)"
|
||||
]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,361 @@
|
||||
{
|
||||
"machine": "nv_resident",
|
||||
"set": "ud_q4_k_xl",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family unsloth --model UD-Q4_K_XL --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model UD-Q4_K_XL --yes --no-tui --context 131072",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure",
|
||||
"file": "qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"gb": 1.2
|
||||
}
|
||||
],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"unsloth",
|
||||
"--model",
|
||||
"UD-Q4_K_XL",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"gpu",
|
||||
"--drafter",
|
||||
"dflash2",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/<STATE>/launch/ud-q4_k_xl-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1230,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.35"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.35)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load unsloth-ud-q4_k_xl",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 1.2,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48",
|
||||
"--batch-mtp"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-q4_k_xl.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"parallel": 3,
|
||||
"aliases": [
|
||||
"unsloth-ud-q4_k_xl"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48",
|
||||
"--batch-mtp",
|
||||
"--batch",
|
||||
"3"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_BATCH_ATTN_ROWS": "1",
|
||||
"STRATA_BATCH_GDN_GROUPS": "1",
|
||||
"STRATA_BATCH_SELF_COMMIT": "1",
|
||||
"STRATA_CKPT_POOL": "1",
|
||||
"STRATA_CKPT_POOL_READY": "4",
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1",
|
||||
"STRATA_QSA_SLOTS": "1",
|
||||
"STRATA_SLOT_CKPT_MOVE": "1",
|
||||
"STRATA_SLOT_COPY_BATCHSYNC": "1",
|
||||
"STRATA_SLOT_COPY_DIRECT": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] Drafting: MTP (DFlash2 isn't available in this mode)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load unsloth-ud-q4_k_xl with Drafting = Auto",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure",
|
||||
"file": "qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"gb": 1.2
|
||||
}
|
||||
],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 1.2,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "draftingFetch",
|
||||
"mode": "auto",
|
||||
"noSplit": false
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-q4_k_xl.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-q4_k_xl"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.q4_0.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.35"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_GDN_CONVL2": "2",
|
||||
"STRATA_GDN_TC": "1"
|
||||
},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.35)"
|
||||
]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,257 @@
|
||||
{
|
||||
"machine": "r9700_linux",
|
||||
"set": "deepseek",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family deepseek41 --model IST-DAS-2.00bpw --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model IST-DAS-2.00bpw --yes --no-tui --context 131072",
|
||||
"fetches": [],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"deepseek41",
|
||||
"--model",
|
||||
"IST-DAS-2.00bpw",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"yes",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/<STATE>/launch/ist-das-2-00bpw-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"engine_ready_s": 3600,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/<STATE>/launch/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load deepseek-v4.1-flash-ist-das-2.00bpw",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": null
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/install/strata-deepseek-v4.1-flash-ist-das-2.00bpw.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"engine_ready_s": 3600,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"aliases": [
|
||||
"deepseek-v4.1-flash-ist-das-2.00bpw"
|
||||
],
|
||||
"api_key": "<KEY>"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/install/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load deepseek-v4.1-flash-ist-das-2.00bpw with Drafting = Auto",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": null
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/install/strata-deepseek-v4.1-flash-ist-das-2.00bpw.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"engine_ready_s": 3600,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"aliases": [
|
||||
"deepseek-v4.1-flash-ist-das-2.00bpw"
|
||||
],
|
||||
"api_key": "<KEY>"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"env_set": {
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/install/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,306 @@
|
||||
{
|
||||
"machine": "r9700_linux",
|
||||
"set": "iq3_s",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family qwen --model IQ3_S --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model IQ3_S --yes --no-tui --context 131072",
|
||||
"fetches": [],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"qwen",
|
||||
"--model",
|
||||
"IQ3_S",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"gpu",
|
||||
"--experimental-speed-projection",
|
||||
"off",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/<STATE>/launch/iq3_s-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1008,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load iq3_s",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/install/strata-iq3_s.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"iq3_s"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load iq3_s with Drafting = Auto",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/install/strata-iq3_s.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"iq3_s"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": []
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,304 @@
|
||||
{
|
||||
"machine": "r9700_linux",
|
||||
"set": "ud_iq4_xs_v31",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family unsloth --model UD-IQ4_XS-v31 --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model UD-IQ4_XS-v31 --yes --no-tui --context 131072",
|
||||
"fetches": [],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"unsloth",
|
||||
"--model",
|
||||
"UD-IQ4_XS-v31",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"gpu",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/<STATE>/launch/ud-iq4_xs-v31-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1089,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load unsloth-ud-iq4_xs-v31",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-iq4_xs-v31.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-iq4_xs-v31"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load unsloth-ud-iq4_xs-v31 with Drafting = Auto",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-iq4_xs-v31.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-iq4_xs-v31"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": []
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,304 @@
|
||||
{
|
||||
"machine": "r9700_linux",
|
||||
"set": "ud_q4_k_xl",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family unsloth --model UD-Q4_K_XL --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model UD-Q4_K_XL --yes --no-tui --context 131072",
|
||||
"fetches": [],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"unsloth",
|
||||
"--model",
|
||||
"UD-Q4_K_XL",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"gpu",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/<STATE>/launch/ud-q4_k_xl-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1230,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load unsloth-ud-q4_k_xl",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-q4_k_xl.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-q4_k_xl"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load unsloth-ud-q4_k_xl with Drafting = Auto",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-q4_k_xl.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-q4_k_xl"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": []
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,269 @@
|
||||
{
|
||||
"machine": "strix_halo_win",
|
||||
"set": "deepseek",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family deepseek41 --model IST-DAS-2.00bpw --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model IST-DAS-2.00bpw --yes --no-tui --context 131072",
|
||||
"fetches": [],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"deepseek41",
|
||||
"--model",
|
||||
"IST-DAS-2.00bpw",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"yes",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/<STATE>/launch/ist-das-2-00bpw-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"engine_ready_s": 3600,
|
||||
"env": {
|
||||
"DS4_ASYNC_T": "4"
|
||||
},
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"env_set": {
|
||||
"DS4_ASYNC_T": "4",
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/<STATE>/launch/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load deepseek-v4.1-flash-ist-das-2.00bpw",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": null
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/install/strata-deepseek-v4.1-flash-ist-das-2.00bpw.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"engine_ready_s": 3600,
|
||||
"env": {
|
||||
"DS4_ASYNC_T": "4"
|
||||
},
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"aliases": [
|
||||
"deepseek-v4.1-flash-ist-das-2.00bpw"
|
||||
],
|
||||
"api_key": "<KEY>"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"env_set": {
|
||||
"DS4_ASYNC_T": "4",
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/install/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load deepseek-v4.1-flash-ist-das-2.00bpw with Drafting = Auto",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": false,
|
||||
"choice": "mtp",
|
||||
"drafter": "mtp",
|
||||
"gb": 0,
|
||||
"line": "Drafting: MTP",
|
||||
"note": "",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": null
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/deepseek/tokenizer",
|
||||
"model_name": "deepseek-v4.1-flash",
|
||||
"arch": "deepseek41",
|
||||
"log": "<ROOT>/install/strata-deepseek-v4.1-flash-ist-das-2.00bpw.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"engine_ready_s": 3600,
|
||||
"env": {
|
||||
"DS4_ASYNC_T": "4"
|
||||
},
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"aliases": [
|
||||
"deepseek-v4.1-flash-ist-das-2.00bpw"
|
||||
],
|
||||
"api_key": "<KEY>"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata-dsv41",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/deepseek",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--eos",
|
||||
"1",
|
||||
"--vision",
|
||||
"on",
|
||||
"--dspark",
|
||||
"auto",
|
||||
"--dspark-conf",
|
||||
"0.65"
|
||||
],
|
||||
"env_set": {
|
||||
"DS4_ASYNC_T": "4",
|
||||
"STRATA_DS4_SLOT_DIR": "<ROOT>/install/ds4-slots",
|
||||
"STRATA_DS4_SLOT_GB": "2",
|
||||
"STRATA_DS4_SLOT_KEEP": "1"
|
||||
},
|
||||
"notes": []
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,347 @@
|
||||
{
|
||||
"machine": "strix_halo_win",
|
||||
"set": "iq3_s",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family qwen --model IQ3_S --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model IQ3_S --yes --no-tui --context 131072",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "hip",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"qwen",
|
||||
"--model",
|
||||
"IQ3_S",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"gpu",
|
||||
"--experimental-speed-projection",
|
||||
"off",
|
||||
"--drafter",
|
||||
"dflash2",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/<STATE>/launch/iq3_s-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1008,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.35"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.35)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load iq3_s",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/install/strata-iq3_s.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"iq3_s"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] Drafting: MTP (the DFlash2 drafter isn't downloaded yet)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load iq3_s with Drafting = Auto",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "hip",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "draftingFetch",
|
||||
"mode": "auto",
|
||||
"noSplit": false
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/iq3_s/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-iq3_s",
|
||||
"log": "<ROOT>/install/strata-iq3_s.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"iq3_s"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/iq3_s",
|
||||
"--native",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
|
||||
"--ple-gguf",
|
||||
"<ROOT>/data/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.35"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.35)"
|
||||
]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,345 @@
|
||||
{
|
||||
"machine": "strix_halo_win",
|
||||
"set": "ud_iq4_xs_v31",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family unsloth --model UD-IQ4_XS-v31 --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model UD-IQ4_XS-v31 --yes --no-tui --context 131072",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "hip",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"unsloth",
|
||||
"--model",
|
||||
"UD-IQ4_XS-v31",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"gpu",
|
||||
"--drafter",
|
||||
"dflash2",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/<STATE>/launch/ud-iq4_xs-v31-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1089,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.35"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.35)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load unsloth-ud-iq4_xs-v31",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-iq4_xs-v31.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-iq4_xs-v31"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] Drafting: MTP (the DFlash2 drafter isn't downloaded yet)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load unsloth-ud-iq4_xs-v31 with Drafting = Auto",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "hip",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "draftingFetch",
|
||||
"mode": "auto",
|
||||
"noSplit": false
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_iq4_xs_v31/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-iq4_xs-v31",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-iq4_xs-v31.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-iq4_xs-v31"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_iq4_xs_v31",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-IQ4_XS-v31/Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00004.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.35"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.35)"
|
||||
]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,345 @@
|
||||
{
|
||||
"machine": "strix_halo_win",
|
||||
"set": "ud_q4_k_xl",
|
||||
"entries": {
|
||||
"setup": {
|
||||
"entry": "setup",
|
||||
"command": "strata setup --family unsloth --model UD-Q4_K_XL --yes",
|
||||
"blocked": {
|
||||
"reason": "setup.py's config generation is one 1,300-line function (run(), setup.py lines ~10100-11400) that, before it writes the run config, probes the GPU with nvidia-smi/rocm-smi, resolves or builds the engine, downloads and SHA-256-checks the model shards, packs the model and (DeepSeek) runs the 16-word smoke test. It cannot be run without a model, an engine and a GPU.",
|
||||
"seam": "a pure function in strata_control (config_for(args, hardware_report, files) -> run config) that run() calls just before write_setup_config(); no behaviour change, a move of the existing dict-building lines. Until then the other entry points start from a stand-in for setup's output (World.base_config)."
|
||||
}
|
||||
},
|
||||
"launch": {
|
||||
"entry": "launch",
|
||||
"command": "strata launch --model UD-Q4_K_XL --yes --no-tui --context 131072",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "hip",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"setup_args": [
|
||||
"--yes",
|
||||
"--no-start",
|
||||
"--family",
|
||||
"unsloth",
|
||||
"--model",
|
||||
"UD-Q4_K_XL",
|
||||
"--context",
|
||||
"131072",
|
||||
"--vision",
|
||||
"gpu",
|
||||
"--drafter",
|
||||
"dflash2",
|
||||
"--data-dir",
|
||||
"<ROOT>/data"
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/<STATE>/launch/ud-q4_k_xl-128k.engine.log",
|
||||
"lib_dirs": [],
|
||||
"port": 8080,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"engine_ready_s": 1230,
|
||||
"args_derived": "=args",
|
||||
"args_written": "=args"
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.35"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.35)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_plain": {
|
||||
"entry": "app_plain",
|
||||
"command": "plain Load unsloth-ud-q4_k_xl",
|
||||
"fetches": [],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": null,
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-q4_k_xl.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-q4_k_xl"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] Drafting: MTP (the DFlash2 drafter isn't downloaded yet)"
|
||||
]
|
||||
}
|
||||
},
|
||||
"app_models": {
|
||||
"entry": "app_models",
|
||||
"command": "Models screen: Load unsloth-ud-q4_k_xl with Drafting = Auto",
|
||||
"fetches": [
|
||||
{
|
||||
"call": "ensure_tuned",
|
||||
"file": "qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"variant": "selq4",
|
||||
"backend": "hip",
|
||||
"gb": 0.34
|
||||
}
|
||||
],
|
||||
"control_calls": [
|
||||
{
|
||||
"call": "draftingPlan",
|
||||
"mode": "auto",
|
||||
"noSplit": false,
|
||||
"result": {
|
||||
"needed": true,
|
||||
"choice": "dflash2",
|
||||
"drafter": "mtp",
|
||||
"gb": 0.34,
|
||||
"line": "Drafting: MTP (the DFlash2 drafter isn't downloaded yet)",
|
||||
"note": "the DFlash2 drafter isn't downloaded yet",
|
||||
"split": false,
|
||||
"mtp_needed": false,
|
||||
"mtp_here": true
|
||||
}
|
||||
},
|
||||
{
|
||||
"call": "draftingFetch",
|
||||
"mode": "auto",
|
||||
"noSplit": false
|
||||
},
|
||||
{
|
||||
"call": "saveProfile"
|
||||
},
|
||||
{
|
||||
"call": "start"
|
||||
}
|
||||
],
|
||||
"config_file": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"args": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--spec",
|
||||
"4",
|
||||
"--spec-min-p",
|
||||
"0.5",
|
||||
"--mtp",
|
||||
"<ROOT>/data/mtp/rt",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48"
|
||||
],
|
||||
"cwd": "<REPO>",
|
||||
"tokenizer": "<ROOT>/data/packs/ud_q4_k_xl/tokenizer",
|
||||
"model_name": "qwen3.8-flash-next-unsloth-ud-q4_k_xl",
|
||||
"log": "<ROOT>/install/strata-unsloth-ud-q4_k_xl.log",
|
||||
"lib_dirs": [],
|
||||
"port": 18500,
|
||||
"backend": "hip",
|
||||
"host": "127.0.0.1",
|
||||
"open_browser": false,
|
||||
"drafting": "auto",
|
||||
"mcp": {},
|
||||
"fit_max_tokens": true,
|
||||
"aliases": [
|
||||
"unsloth-ud-q4_k_xl"
|
||||
],
|
||||
"api_key": "<KEY>",
|
||||
"engine_ready_s": 946
|
||||
},
|
||||
"engine": {
|
||||
"exe": "<ROOT>/engine/strata",
|
||||
"argv": [
|
||||
"--pack",
|
||||
"<ROOT>/data/packs/ud_q4_k_xl",
|
||||
"--native",
|
||||
"<ROOT>/data/models/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00003.gguf",
|
||||
"--expert-profile",
|
||||
"<REPO>/data/expert-profile.bin",
|
||||
"--expert-cache",
|
||||
"auto",
|
||||
"--prefill",
|
||||
"auto",
|
||||
"--max-context",
|
||||
"131072",
|
||||
"--kv",
|
||||
"int8",
|
||||
"--resident-budget-gib",
|
||||
"48",
|
||||
"--draft",
|
||||
"dflash2",
|
||||
"--draft-model",
|
||||
"<ROOT>/data/dflash2/qwen3.8-flash-next-dflash2.strata1.q4_0-selq4.pack",
|
||||
"--draft-fallback",
|
||||
"none",
|
||||
"--mtp-draft-vocab",
|
||||
"<ROOT>/data/mtp/rt/draft_vocab.bin",
|
||||
"--spec",
|
||||
"6",
|
||||
"--draft-conf",
|
||||
"0.35"
|
||||
],
|
||||
"env_set": {},
|
||||
"notes": [
|
||||
"[strata] drafting: DFlash2 alone (--draft dflash2, --draft-conf 0.35)"
|
||||
]
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
Reference in new issue
Block a user