Niko1221 cc3f9142e3 Auto on the 12 GB / streaming CUDA class: IQ3_S starts the MTP, UD-IQ4_XS-v31 starts DFlash2 with the tuned pack at the class's own 0.8 ...

drafting.CLASS_DEFAULTS: one removable entry per set, read from the signals the confidence already reads (a CUDA card whose engine slots are under the last-but-one CONF_STEPS step).
IQ3_S: on P2003 the file cache makes the MTP faster than DFlash2 by 12-44% (RTX2 t4b). UD-IQ4_XS-v31: DFlash2 beats the MTP by about 6.5% (t8, 3 starts; DF-VAL forced-fast tuned 0.8 x1.037).
The v31 entry DEPENDS ON RTX2's FILE-TIER FIX (unbuffered reads) BEING IN THE SAME RELEASE; without it remove that entry.
Anything chosen by name wins. Resident CUDA, Strix, gfx1201, Arc and Metal entries and the conf tables are unchanged. The sm_86 support line reads the set: UD-IQ4_XS-v31 gets the existing
'faster on some kinds of text and slower on others' line, every other model keeps 'slower'.

(cherry picked from commit 42412d42ff835d6a66a78fbf6d119911908eda1b)

2026-11-06 15:47:42 +01:00

..

__init__.py

launch: settings, catalog (setup.py tables via strata_mcp, fitOn port), presets store, access, runner, flags, connect, harness drafts

2026-10-16 23:25:34 +02:00

access.py

launch: a damaged profile no longer lends its neighbour's config or mints a new API key (#4395, #4396)

2026-11-04 10:36:05 +01:00

adopt.py

launch: Unsloth's sets are written for an Intel Arc card (adopt.FAMILIES), without --ple-gguf; test

2026-11-05 08:30:25 +01:00

app.py

One drafter per run: DFlash2 is the primary drafter where the engine has it (CUDA, HIP, Metal), alone and with --draft-fallback none; the MTP is chosen alone with no DFlash2 option

2026-11-02 01:31:26 +01:00

catalog.py

catalog: Unsloth v3.1 Qwen3.8-Flash-Next sets (UD-IQ1_M, UD-Q2_0, UD-IQ2_XS, UD-Q2_K_XL, UD-IQ3_XXS, UD-Q3_K_XL, UD-IQ4_XS-v31) at d9620119 with a per-model revision; UD-Q4_K_XL's four-shard layout accepted from local files; manual-load refusal lifted (UD-IQ1_S stays refused); mac_fit switch for the Q5_0 down sets; header-only fixtures and tests (about lines pending wording)

2026-11-05 07:23:41 +01:00

check.py

Calls to this computer's own servers skip the system proxy (tuner, tune session, launch self-check)

2026-10-20 13:58:47 +02:00

cmd.py

launch: 'A new API key was made' is said only once the run config holding the key has been written (never in the plan or before a refusal)

2026-11-04 17:19:20 +01:00

harness.py

connect: the output cap grows with the window, and a rewrite keeps a cap the person set (#5006; #4528 checked, no change)

2026-11-04 10:56:16 +01:00

library.py

[Other] --data-dir never moves the model library without a choice

2026-10-19 20:37:29 +02:00

modelstate.py

launcher: draft layer also found in --data-dir, split engine counts free VRAM, plan card numbers are the --gpu numbers

2026-11-04 07:11:31 +01:00

monitor_ext.py

Tests follow the new rules: a model beside an update is refused before any download; the launch no longer upgrades DeepSeek; the monitor line

2026-11-01 13:36:10 +01:00

params.py

Setup, Control and Terminal: a child's non-UTF-8 output (GBK adapter names) no longer kills Step 1 or launch (#4675)

2026-10-30 18:44:41 +01:00

runner.py

Auto on the 12 GB / streaming CUDA class: IQ3_S starts the MTP, UD-IQ4_XS-v31 starts DFlash2 with the tuned pack at the class's own 0.8

2026-11-06 15:47:42 +01:00

screens.py

Drafting follows the model: the Terminal goes back to Auto when the model changes to one without a Drafting choice (DeepSeek, Coder, Swift, NVFP4) and a launch writes none into a DeepSeek run config; the app's Models page resets to Auto for a model that is not installed; the Q4_0-selector pack is made only where the installed engine reads it (never for the SYCL engine); tests

2026-11-02 19:37:22 +01:00

settings.py

launch: 'A new API key was made' is said only once the run config holding the key has been written (never in the plan or before a refusal)

2026-11-04 17:19:20 +01:00

store.py

Save and launch saves the preset only after the restart question is answered Restart; the question names the running model and port

2026-10-23 03:27:48 +02:00

widgets.py

Terminal wizard: the Save choices stay on screen at 60x20 and 40x15, the Advanced command has its own scroll area, key hints wrap and say what Enter does; request filter explains itself and matches LIVE/DONE; plain-word setting descriptions; tuning default explained; monitor bundle carries nav.py (T09, T11)

2026-10-18 23:23:09 +02:00