Niko1221 6f7f680529 verify: the all-resident graph runs only while every expert is in VRAM; a -1 raises a plan error flag (#871) ...

A prompt loan, a VRAM shrink or an adaptive swap marks experts -1 in the residency table.  The zero-doorbell graph
planned from the device table and returned silently on a -1, so the window ran a stale plan (!!!! at 100% residency).
refresh_ar() checks the host table before each window and runs the doorbell graph (its own captured variant) while any
expert is out; resident_plan also writes a mapped error flag the host checks after the window.

2026-10-05 11:03:50 +02:00

..

conversation_buffer.hpp

Bound incremental snapshot fragmentation and preserve cache admission

2026-09-30 16:31:46 +02:00

conversation_cache.hpp

cache: preserve checkpoints when split allocation fails

2026-10-05 10:00:33 +02:00

conversation_memory.hpp

Integrate validated conversation snapshot core with upstream 0.1.25

2026-09-30 16:31:44 +02:00

conversation_snapshot.hpp

Parking with a layer split: the draft ring once, every stage's K/V reused, checkpoints moved not copied

2026-10-03 17:03:02 +02:00

coupled_draft.hpp

feat(mtp): add --coupled-draft and --no-coupled-draft CLI options for coupled speculative sampling

2026-10-01 08:15:27 +02:00

device.hpp

strata-device --list-devices, and HIP's numbering for AMD cards on Windows (#325)

2026-10-01 20:19:56 +02:00

emulate.hpp

Start: name the GPU and stop at once when the build has no code for it; STRATA_EMULATE_CC for tests

2026-10-01 08:11:23 +02:00

expert_cache.hpp

core: optimize remote expert helper decode

2026-10-03 23:33:07 +02:00

expert_source.hpp

Merge branch 'f-139-prs' into f-139-rel

2026-10-03 20:04:42 +02:00

graph.hpp

Strata: Qwen3.8-Flash-Next on a 12-24 GB NVIDIA GPU + 64 GB RAM, one-click install

2026-09-24 18:42:20 +02:00

hit_hook.hpp

Strata: Qwen3.8-Flash-Next on a 12-24 GB NVIDIA GPU + 64 GB RAM, one-click install

2026-09-24 18:42:20 +02:00

layer.hpp

gfx906: route the verify window's tokens in one BF16 projection + one top-k (bitwise the per-token calls)

2026-10-03 23:33:00 +02:00

layout.hpp

Strata: Qwen3.8-Flash-Next on a 12-24 GB NVIDIA GPU + 64 GB RAM, one-click install

2026-09-24 18:42:20 +02:00

mtp.hpp

perf(cuda,decode): zero-doorbell verify graph, sub-warp expert packing, shared-mem staging, and batched PLE/MTP

2026-10-03 23:27:36 +02:00

native_dense.hpp

Merge PR #559 (blange48, crazyaimachine): batch slots, pipelined layer split, parking with a split

2026-10-03 23:29:40 +02:00

native_head.hpp

Split GGUF artifacts: every shard found (a missing one is an error), tensors from their owning shard, per-role expert files (native_experts.txt v4)

2026-09-30 22:30:10 +02:00

on_device.hpp

Layer split v1 (step 4): the second stage on its own GPU

2026-09-28 23:37:29 +02:00

peer_experts.hpp

multi-GPU: a second GPU as an opt-in expert-cache tier (--peer-device N)

2026-10-02 22:04:42 +02:00

pinned.hpp

File tier: decide unbuffered reads from the expert bytes it reads, again once the RAM copy is built (#577)

2026-10-03 18:12:53 +02:00

progress.hpp

Stall report: name the prompt chunk and layer; mark a finished verify window (#251)

2026-09-30 21:03:36 +02:00

remote_expert_opt.hpp

core: optimize remote expert helper decode

2026-10-03 23:33:07 +02:00

remote_experts.hpp

core: optimize remote expert helper decode

2026-10-03 23:33:07 +02:00

session.hpp

cpu: preserve full thread affinity on Windows and Linux (#626)

2026-10-04 01:29:24 +02:00

verify.hpp

verify: the all-resident graph runs only while every expert is in VRAM; a -1 raises a plan error flag (#871)

2026-10-05 11:03:50 +02:00

weights.hpp

native packs: a BF16 (--compat-bf16) PLE key stays in the arena and is not replaced by the GGUF key (#326)

2026-10-01 08:29:03 +02:00