FreeToken ROCm Watch — notification channel

[Castoff995]

Этот PR служит постоянным каналом уведомлений для автоматической проверки GitHub каждые 2 часа. Его не нужно ревьюить, мержить или закрывать, пока слежка нужна.

Высший приоритет

• FlashML-org/FreeToken PR feat(rocm): add RDNA3 and RDNA4 runtime foundation FlashML-org/FreeToken#132–feat(rocm): serve on AMD GPUs through the HIP toolchain FlashML-org/FreeToken#137: ROCm/RDNA3/RDNA4 foundation, TVM-FFI HIP JIT, backend gating, RCCL, GGUF RDNA4 и альтернативный ROCm port.
• Issues ROCm RDNA4 gfx1201 Ported. Windows only for now. Test please FlashML-org/FreeToken#82, Native Windows is classified as pin-uncapped, so #112's per-layer residency never auto-engages FlashML-org/FreeToken#120, Windows/ROCm on RX 9060 XT (gfx1200): packed GGUF working end-to-end (dense + MoE), root cause of the RDNA4 offload-decode TDR, benchmarks vs llama.cpp FlashML-org/FreeToken#122 и roadmap FreeToken Roadmap (2026) FlashML-org/FreeToken#79.
• Новые upstream issues/PR с AMD, ROCm, HIP, RDNA4, gfx1200, gfx1201, RX 9070, Windows, Triton, TVM-FFI, pinned/mapped host memory, hipHostGetDevicePointer, launch_pdl, clang-cl, MoE/offload.

Рабочие форки

• PialGhosh2233/FreeToken-rocm-gfx1200:gfx1200-moe-gguf.
• Maxritz/FreeToken-rocm-test:main.

Зависимости

• ROCm/TheRock — Windows/RDNA4/gfx1201 wheels, PyTorch/HIP regressions and releases.
• triton-lang/triton-windows — Windows AMD/HIP support and releases.
• apache/tvm-ffi — Windows/HIP JIT and cache/toolchain changes.

Устойчивость модели

Также отслеживаются изменения, способные повлиять на первый токен и практический перевод: Qwen3.6/NVFP4, MoE offload, hybrid_radix, pinned/pageable residency, mapped device pointers, long-prefill/KV-cache hangs and engine crashes.

Уведомление публикуется только при существенном изменении: новый релевантный issue/PR, новый commit в ключевом PR/форке, изменение draft/merge/CI/state, новый релиз либо подтверждённый gfx1201/Windows результат. При отсутствии новостей комментариев не будет.

[github-actions]

@Castoff995 FreeToken ROCm Watch запущен. Проверка выполняется каждые 2 часа; при отсутствии существенных изменений комментариев не будет.

Текущий P0 baseline:

• PR feat(rocm): add RDNA3 and RDNA4 runtime foundation FlashML-org/FreeToken#132: open, head 45417560, checks pending — feat(rocm): add RDNA3 and RDNA4 runtime foundation
• PR fix(rocm): make TVM-FFI index and store JIT kernels portable to HIP FlashML-org/FreeToken#133: open / draft, head 89ad7ab0, checks pending — fix(rocm): make TVM-FFI index and store JIT kernels portable to HIP
• PR fix(rocm): gate CUDA-only optional backends on ROCm FlashML-org/FreeToken#134: open / draft, head 9a523dc8, checks pending — fix(rocm): gate CUDA-only optional backends on ROCm
• PR fix(rocm): route tensor parallel communication through RCCL FlashML-org/FreeToken#135: open / draft, head 9d5e3fc8, checks pending — fix(rocm): route tensor parallel communication through RCCL
• PR feat(rocm): enable native GGUF kernels on ROCm (validated on RDNA4) FlashML-org/FreeToken#136: open / draft, head 49591978, checks pending — feat(rocm): enable native GGUF kernels on RDNA4
• PR feat(rocm): serve on AMD GPUs through the HIP toolchain FlashML-org/FreeToken#137: open, head 7e1d7421, checks pending — feat(rocm): serve on AMD GPUs through the HIP toolchain

Ключевые issues: FreeToken Roadmap (2026), ROCm RDNA4 gfx1201 Ported. Windows only for now. Test please, Native Windows is classified as pin-uncapped, so #112's per-layer residency never auto-engages, Windows/ROCm on RX 9060 XT (gfx1200): packed GGUF working end-to-end (dense + MoE), root cause of the RDNA4 offload-decode TDR, benchmarks vs llama.cpp.

Дополнительно отслеживаются fork commits, ROCm/TheRock, Triton Windows, TVM-FFI, PyTorch gfx1201, Qwen3.6/NVFP4, mapped host pointers, pin-budget, long-prefill и offload stability.

[github-actions]

{"branches":{"FlashML-org/FreeToken:main":{"branch":"main","date":"2026-09-04T06:52:33Z","message":"ci: publish engine-.json manifests alongside the nightly wheels (FlashML-org#377)","repo":"FlashML-org/FreeToken","sha":"af71ba43206e124f5ff6419b47ee36c6e9981078","url":"FlashML-org@af71ba43206e124f5ff6419b47ee36c6e9981078"},"Maxritz/FreeToken-rocm-test:main":{"branch":"main","date":"2026-08-26T08:26:56Z","message":"Merge pull request Maxritz#2 from PialGhosh2233/main","repo":"Maxritz/FreeToken-rocm-test","sha":"89f14790de7bbc69dd61a999ee76b86ba8316f26","url":"Maxritz@89f14790de7bbc69dd61a999ee76b86ba8316f26"},"PialGhosh2233/FreeToken-rocm-gfx1200:gfx1200-moe-gguf":{"branch":"gfx1200-moe-gguf","date":"2026-08-26T07:51:22Z","message":"docs: Gemma-4-26B-A4B QAT q4_0 GGUF results on gfx1200","repo":"PialGhosh2233/FreeToken-rocm-gfx1200","sha":"a338f49b93e2f098eeb3aa83423c9c5386c53ac4","url":"PialGhosh2233@a338f49b93e2f098eeb3aa83423c9c5386c53ac4"}},"discovered":{"FlashML-org/FreeToken#104%22:%7B%22body_hash%22:%220ba9970c69c92be9%22,%22comments%22:16,%22kind%22:%22pr%22,%22number%22:104,%22repo%22:%22FlashML-org/FreeToken%22,%22state%22:%22open%22,%22title%22:%22feat(models): support TP for qwen3_5_moe","updated_at":"2026-09-04T15:09:13Z","url":"https://github.com/FlashML-org/FreeToken/pull/104"},"FlashML-org/FreeToken#108":{"body_hash":"d12d8bede30d7b6c","comments":2,"kind":"pr","number":108,"repo":"FlashML-org/FreeToken","state":"open","title":"feat(server): Add OpenAI /v1/models/{model_id} endpoint","updated_at":"2026-09-05T04:24:16Z","url":"https://github.com/FlashML-org/FreeToken/pull/108"},"FlashML-org/FreeToken#16":{"body_hash":"fb266d8bbe40557f","comments":1,"kind":"issue","number":16,"repo":"FlashML-org/FreeToken","state":"closed","title":"NotImplementedError: GPUs on both sides of the sm_89 fp8 boundary in one process: the host-side e4m3 convention is process-global","updated_at":"2026-09-04T09:08:50Z","url":"https://github.com/FlashML-org/FreeToken/issues/16"},"FlashML-org/FreeToken#169":{"body_hash":"742a514547b6b67c","comments":1,"kind":"pr","number":169,"repo":"FlashML-org/FreeToken","state":"open","title":"fix(engine): load prefill triton kernels at startup, not mid-request","updated_at":"2026-09-05T04:25:08Z","url":"https://github.com/FlashML-org/FreeToken/pull/169"},"FlashML-org/FreeToken#207":{"body_hash":"9d6c44260cb6edc1","comments":3,"kind":"issue","number":207,"repo":"FlashML-org/FreeToken","state":"closed","title":"[Linux] AppImage blank window / WebKitWebProcess abort: "Could not create default EGL display: EGL_BAD_PARAMETER" (bundled libwayland-client on Mesa 25+ hosts)","updated_at":"2026-09-08T03:31:24Z","url":"https://github.com/FlashML-org/FreeToken/issues/207"},"FlashML-org/FreeToken#217":{"body_hash":"0e0dbaaf50222ee4","comments":3,"kind":"pr","number":217,"repo":"FlashML-org/FreeToken","state":"open","title":"feat(rocm): AMD ROCm runtime for RDNA3/4 with native Qwen3.5-MoE GGUF decode parity","updated_at":"2026-09-05T19:32:42Z","url":"https://github.com/FlashML-org/FreeToken/pull/217"},"FlashML-org/FreeToken#227":{"body_hash":"ec213ea17037ea8c","comments":12,"kind":"issue","number":227,"repo":"FlashML-org/FreeToken","state":"closed","title":"[Bug]: RuntimeError: Rowwise scaling is not currently supported on your device","updated_at":"2026-09-08T04:20:41Z","url":"https://github.com/FlashML-org/FreeToken/issues/227"},"FlashML-org/FreeToken#248":{"body_hash":"89bb063d23656e84","comments":2,"kind":"issue","number":248,"repo":"FlashML-org/FreeToken","state":"closed","title":"Two blocking bugs in Windows Desktop on consumer Blackwell (RTX 50): FP8 row-wise gate crash + Mistral tokenizer empty Chinese output","updated_at":"2026-09-08T04:21:55Z","url":"https://github.com/FlashML-org/FreeToken/issues/248"},"FlashML-org/FreeToken#252":{"body_hash":"01ce46ac32f028fd","comments":4,"kind":"issue","number":252,"repo":"FlashML-org/FreeToken","state":"open","title":"[Feature Request] Support compressed-tensors per-channel FP8 checkpoints for Qwen3.5 MoE","updated_at":"2026-09-08T02:26:17Z","url":"https://github.com/FlashML-org/FreeToken/issues/252"},"FlashML-org/FreeToken#260":{"body_hash":"e0794d434a8a819d","comments":5,"kind":"pr","number":260,"repo":"FlashML-org/FreeToken","state":"open","title":"ROCm: validate native gfx1151 serving on Strix Halo","updated_at":"2026-09-05T22:42:23Z","url":"https://github.com/FlashML-org/FreeToken/pull/260"},"FlashML-org/FreeToken#268":{"body_hash":"90e395f1fec25670","comments":10,"kind":"pr","number":268,"repo":"FlashML-org/FreeToken","state":"open","title":"feat(kvcache): sub-byte Q4_0 and Q6_0 KV cache quantization","updated_at":"2026-09-04T16:19:48Z","url":"https://github.com/FlashML-org/FreeToken/pull/268"},"FlashML-org/FreeToken#275":{"body_hash":"2325c00fd601603b","comments":4,"kind":"pr","number":275,"repo":"FlashML-org/FreeToken","state":"open","title":"fix: [BUG] qwen3_5 dense: mixed-precision NVFP4 crashes in ct_bf16_fuse (Float8 × BFloat16 promotion)","updated_at":"2026-09-08T01:19:55Z","url":"https://github.com/FlashML-org/FreeToken/pull/275"},"FlashML-org/FreeToken#280":{"body_hash":"6abdea5758e2a7f7","comments":1,"kind":"issue","number":280,"repo":"FlashML-org/FreeToken","state":"open","title":"Feature Request: opt-in FP8 KV-cache support for long-context MoE inference","updated_at":"2026-09-05T23:32:30Z","url":"https://github.com/FlashML-org/FreeToken/issues/280"},"FlashML-org/FreeToken#287":{"body_hash":"53577e1bbcf544bd","comments":2,"kind":"pr","number":287,"repo":"FlashML-org/FreeToken","state":"closed","title":"fix(scheduler): hybrid radix hit corruption — GDN snapshot copy-on-donate + admission barriers","updated_at":"2026-09-05T10:13:20Z","url":"https://github.com/FlashML-org/FreeToken/pull/287"},"FlashML-org/FreeToken#298":{"body_hash":"82f5c00b90b8d5d3","comments":0,"kind":"pr","number":298,"repo":"FlashML-org/FreeToken","state":"open","title":"Support poolside/Laguna-S-2.1-NVFP4, and fix serving from source on Windows","updated_at":"2026-09-05T18:31:59Z","url":"https://github.com/FlashML-org/FreeToken/pull/298"},"FlashML-org/FreeToken#299":{"body_hash":"fe817aa800d84eb9","comments":7,"kind":"issue","number":299,"repo":"FlashML-org/FreeToken","state":"open","title":"FreeToken-Intel: 35B MoE, live tokens off an Arc Pro B70 (SYCL / XPU port)","updated_at":"2026-09-04T23:41:52Z","url":"https://github.com/FlashML-org/FreeToken/issues/299"},"FlashML-org/FreeToken#301":{"body_hash":"40475c88b787fc6e","comments":2,"kind":"issue","number":301,"repo":"FlashML-org/FreeToken","state":"open","title":"fix(server): allow overriding the distributed port","updated_at":"2026-09-06T12:13:08Z","url":"https://github.com/FlashML-org/FreeToken/issues/301"},"FlashML-org/FreeToken#302":{"body_hash":"53ce92fba4ae82bc","comments":3,"kind":"issue","number":302,"repo":"FlashML-org/FreeToken","state":"open","title":"feat(server): add a server-side thinking override","updated_at":"2026-09-06T11:55:42Z","url":"https://github.com/FlashML-org/FreeToken/issues/302"},"FlashML-org/FreeToken#306":{"body_hash":"5f4417a782901609","comments":2,"kind":"issue","number":306,"repo":"FlashML-org/FreeToken","state":"open","title":"Long-context decode: client wall-clock is ~half the scheduler gen throughput","updated_at":"2026-09-05T23:29:09Z","url":"https://github.com/FlashML-org/FreeToken/issues/306"},"FlashML-org/FreeToken#307":{"body_hash":"0c7eb3672129dec5","comments":1,"kind":"issue","number":307,"repo":"FlashML-org/FreeToken","state":"open","title":"temperature 0 is not reproducible with thinking on (identical requests diverge on near-ties)","updated_at":"2026-09-05T17:21:03Z","url":"https://github.com/FlashML-org/FreeToken/issues/307"},"FlashML-org/FreeToken#320":{"body_hash":"dc06711149928e6d","comments":2,"kind":"pr","number":320,"repo":"FlashML-org/FreeToken","state":"open","title":"qwen4_exp: load modelopt MIXED_PRECISION (NVFP4 experts + block-FP8 dense) checkpoints","updated_at":"2026-09-04T22:14:01Z","url":"https://github.com/FlashML-org/FreeToken/pull/320"},"FlashML-org/FreeToken#338":{"body_hash":"2ba2e586b8540644","comments":4,"kind":"pr","number":338,"repo":"FlashML-org/FreeToken","state":"open","title":"perf(ple): fuse the n-gram row-id hash into one Triton kernel","updated_at":"2026-09-04T16:38:26Z","url":"https://github.com/FlashML-org/FreeToken/pull/338"},"FlashML-org/FreeToken#339":{"body_hash":"7d0f17426b165cb7","comments":4,"kind":"pr","number":339,"repo":"FlashML-org/FreeToken","state":"open","title":"fix(kernels): drop the D2H sync from the varlen GDN/KDA prefill conv","updated_at":"2026-09-04T15:00:40Z","url":"https://github.com/FlashML-org/FreeToken/pull/339"},"FlashML-org/FreeToken#340":{"body_hash":"4c2febbad30c626f","comments":1,"kind":"pr","number":340,"repo":"FlashML-org/FreeToken","state":"open","title":"fix(engine): reserve the pool's dummy page in the --moe-cache-auto KV floor","updated_at":"2026-09-04T15:00:41Z","url":"https://github.com/FlashML-org/FreeToken/pull/340"},"FlashML-org/FreeToken#342":{"body_hash":"52be93d51888c01f","comments":1,"kind":"pr","number":342,"repo":"FlashML-org/FreeToken","state":"open","title":"qwen3_5_moe: run lm_head on sampled rows only (fixes 32 GB first-prefill OOM)","updated_at":"2026-09-04T15:50:15Z","url":"https://github.com/FlashML-org/FreeToken/pull/342"},"FlashML-org/FreeToken#35":{"body_hash":"496be1e928015afc","comments":3,"kind":"issue","number":35,"repo":"FlashML-org/FreeToken","state":"closed","title":"FreeToken Engine Install Fails in Windows with RTX4060","updated_at":"2026-09-08T03:33:20Z","url":"https://github.com/FlashML-org/FreeToken/issues/35"},"FlashML-org/FreeToken#354":{"body_hash":"8345cf73d9d7fb75","comments":14,"kind":"pr","number":354,"repo":"FlashML-org/FreeToken","state":"open","title":"feat(kvcache): store the KV cache as fp8 e4m3 codes (--kv-cache-dtype…","updated_at":"2026-09-07T23:49:19Z","url":"https://github.com/FlashML-org/FreeToken/pull/354"},"FlashML-org/FreeToken#362":{"body_hash":"23cc7b624683db16","comments":2,"kind":"issue","number":362,"repo":"FlashML-org/FreeToken","state":"open","title":"[Feature Request]: Support EXL3 Format","updated_at":"2026-09-07T18:25:15Z","url":"https://github.com/FlashML-org/FreeToken/issues/362"},"FlashML-org/FreeToken#365":{"body_hash":"3831930353df160b","comments":2,"kind":"issue","number":365,"repo":"FlashML-org/FreeToken","state":"closed","title":"Qwen3.5-122B-A10B-NVFP4","updated_at":"2026-09-07T13:58:39Z","url":"https://github.com/FlashML-org/FreeToken/issues/365"},"FlashML-org/FreeToken#371":{"body_hash":"7538342caf26a9eb","comments":1,"kind":"issue","number":371,"repo":"FlashML-org/FreeToken","state":"open","title":"TP > 1: every rank keeps torch's default intra-op thread count, so a multi-rank weight load busy-waits itself to a standstill on a small-core host (19 min -> 61 s)","updated_at":"2026-09-05T23:29:52Z","url":"https://github.com/FlashML-org/FreeToken/issues/371"},"FlashML-org/FreeToken#381":{"body_hash":"7b7953cbc7889e51","comments":0,"kind":"issue","number":381,"repo":"FlashML-org/FreeToken","state":"open","title":"[Bug]: qwen3_5_moe fails to convert/load compressed-tensors NVFP4 checkpoints with model.language_model prefix","updated_at":"2026-09-04T10:34:22Z","url":"https://github.com/FlashML-org/FreeToken/issues/381"},"FlashML-org/FreeToken#383":{"body_hash":"7b7e372a1d279391","comments":1,"kind":"issue","number":383,"repo":"FlashML-org/FreeToken","state":"open","title":"--moe-cache-auto ignores --num-tokens / --num-pages: expert fill overruns the budget and the KV pool OOMs at boot","updated_at":"2026-09-04T19:28:00Z","url":"https://github.com/FlashML-org/FreeToken/issues/383"},"FlashML-org/FreeToken#384":{"body_hash":"a5cfef075e1a00c8","comments":0,"kind":"issue","number":384,"repo":"FlashML-org/FreeToken","state":"open","title":"Build on Windows?","updated_at":"2026-09-04T15:32:45Z","url":"https://github.com/FlashML-org/FreeToken/issues/384"},"FlashML-org/FreeToken#385":{"body_hash":"387898d8b1fab043","comments":1,"kind":"pr","number":385,"repo":"FlashML-org/FreeToken","state":"open","title":"feat(qwen4_exp): tensor parallelism for Qwen3.8-Flash-Next (offload backend)","updated_at":"2026-09-04T19:20:01Z","url":"https://github.com/FlashML-org/FreeToken/pull/385"},"FlashML-org/FreeToken#387":{"body_hash":"351ce6d8d13b27f5","comments":0,"kind":"issue","number":387,"repo":"FlashML-org/FreeToken","state":"open","title":"Desktop engine installer ignores existing FreeToken engine path and always installs a new venv in AppData","updated_at":"2026-09-04T18:45:08Z","url":"https://github.com/FlashML-org/FreeToken/issues/387"},"FlashML-org/FreeToken#389":{"body_hash":"dcccc6e69f4d67e3","comments":2,"kind":"pr","number":389,"repo":"FlashML-org/FreeToken","state":"open","title":"feat(qwen4_exp): load-time per-tensor FP8 dense projections (W8A8 via _scaled_mm)","updated_at":"2026-09-05T21:48:07Z","url":"https://github.com/FlashML-org/FreeToken/pull/389"},"FlashML-org/FreeToken#390":{"body_hash":"a2503c9f178eed1b","comments":3,"kind":"pr","number":390,"repo":"FlashML-org/FreeToken","state":"open","title":"fix(models): better support for mixed-precision compressed-tensors NVFP4","updated_at":"2026-09-08T07:33:59Z","url":"https://github.com/FlashML-org/FreeToken/pull/390"},"FlashML-org/FreeToken#392":{"body_hash":"2965b722053667b3","comments":3,"kind":"pr","number":392,"repo":"FlashML-org/FreeToken","state":"open","title":"qwen4_exp: serve the block-FP8 dense projections natively (+25% decode)","updated_at":"2026-09-05T23:20:02Z","url":"https://github.com/FlashML-org/FreeToken/pull/392"},"FlashML-org/FreeToken#393":{"body_hash":"cdaa24a2eb622868","comments":0,"kind":"pr","number":393,"repo":"FlashML-org/FreeToken","state":"open","title":"feat(server): OpenAI protocol coverage -- sampling extras, n, echo/suffix, usage details, tokenize/detokenize/metrics","updated_at":"2026-09-05T13:00:52Z","url":"https://github.com/FlashML-org/FreeToken/pull/393"},"FlashML-org/FreeToken#394":{"body_hash":"a78be87d98a48a00","comments":1,"kind":"issue","number":394,"repo":"FlashML-org/FreeToken","state":"open","title":"Support official nvidia/Qwen3.8-Flash-Next-NVFP4 model","updated_at":"2026-09-05T23:26:11Z","url":"https://github.com/FlashML-org/FreeToken/issues/394"},"FlashML-org/FreeToken#395":{"body_hash":"06db136ff57c621d","comments":1,"kind":"issue","number":395,"repo":"FlashML-org/FreeToken","state":"open","title":"[Bug] --max-output-tokens is ignored by /v1/chat/completions and /v1/messages (only /v1/responses honours it)","updated_at":"2026-09-05T23:31:37Z","url":"https://github.com/FlashML-org/FreeToken/issues/395"},"FlashML-org/FreeToken#396":{"body_hash":"5c11b3569ee092da","comments":0,"kind":"issue","number":396,"repo":"FlashML-org/FreeToken","state":"open","title":"[Feature] int4 routed experts: AWQ -> Q4_1 and compressed-tensors pack-quantized -> Q4_0, with no new CUDA kernel (offering a PR)","updated_at":"2026-09-06T05:40:49Z","url":"https://github.com/FlashML-org/FreeToken/issues/396"},"FlashML-org/FreeToken#397":{"body_hash":"33f938edd354e81a","comments":0,"kind":"issue","number":397,"repo":"FlashML-org/FreeToken","state":"open","title":"5090+256G内存无法运行红帽的GLM5.3FlashNVFP4","updated_at":"2026-09-05T19:21:52Z","url":"https://github.com/FlashML-org/FreeToken/issues/397"},"FlashML-org/FreeToken#398":{"body_hash":"f27e8711e749cf12","comments":0,"kind":"pr","number":398,"repo":"FlashML-org/FreeToken","state":"open","title":"feat(kvcache): add ISO3/ISO4 KV-cache quantization (--kv-cache-iso)","updated_at":"2026-09-05T20:36:20Z","url":"https://github.com/FlashML-org/FreeToken/pull/398"},"FlashML-org/FreeToken#399":{"body_hash":"227a0901fd907e75","comments":0,"kind":"pr","number":399,"repo":"FlashML-org/FreeToken","state":"open","title":"fix(cpu-moe): honor padded fp8 scale strides, add avx512f tier and float64 parity (builds on FlashML-org#36)","updated_at":"2026-09-06T00:21:38Z","url":"https://github.com/FlashML-org/FreeToken/pull/399"},"FlashML-org/FreeToken#401":{"body_hash":"b1163ed10ceb8548","comments":0,"kind":"issue","number":401,"repo":"FlashML-org/FreeToken","state":"open","title":"--moe-cache-auto leaves no room for prefill: boots and serves decode, then the first 8k prefill OOMs and kills the worker (NVFP4 dense)","updated_at":"2026-09-06T04:11:21Z","url":"https://github.com/FlashML-org/FreeToken/issues/401"},"FlashML-org/FreeToken#403":{"body_hash":"76bd4f986529962a","comments":2,"kind":"issue","number":403,"repo":"FlashML-org/FreeToken","state":"closed","title":"prompt is too long: 16535 tokens > 8221 maximum","updated_at":"2026-09-07T07:42:14Z","url":"https://github.com/FlashML-org/FreeToken/issues/403"},"FlashML-org/FreeToken#404":{"body_hash":"413b22ce94b35bb6","comments":0,"kind":"pr","number":404,"repo":"FlashML-org/FreeToken","state":"open","title":"fix(models): support nvidia/Qwen3.8-Flash-Next-NVFP4 in qwen4_exp","updated_at":"2026-09-07T02:31:21Z","url":"https://github.com/FlashML-org/FreeToken/pull/404"},"FlashML-org/FreeToken#405":{"body_hash":"457abec0397d19e5","comments":0,"kind":"pr","number":405,"repo":"FlashML-org/FreeToken","state":"open","title":"fix(models): serve qwen4_exp FTW checkpoints with PLE streamed from source","updated_at":"2026-09-07T03:35:03Z","url":"https://github.com/FlashML-org/FreeToken/pull/405"},"FlashML-org/FreeToken#406":{"body_hash":"caec58fadfcf0f28","comments":0,"kind":"pr","number":406,"repo":"FlashML-org/FreeToken","state":"open","title":"docs(models): qualify full GLM-5.3 on H200 Serverless","updated_at":"2026-09-07T21:15:09Z","url":"https://github.com/FlashML-org/FreeToken/pull/406"},"FlashML-org/FreeToken#407":{"body_hash":"75c7c60da2dac36a","comments":1,"kind":"issue","number":407,"repo":"FlashML-org/FreeToken","state":"open","title":"Qwen3.6‑35B‑A3B‑Uncensored‑HauhauCS‑Aggressive GGUF 加载报错 qwen35moe architecture not supported","updated_at":"2026-09-07T11:03:43Z","url":"https://github.com/FlashML-org/FreeToken/issues/407"},"FlashML-org/FreeToken#408":{"body_hash":"c59b1167f4e8a6b6","comments":0,"kind":"pr","number":408,"repo":"FlashML-org/FreeToken","state":"open","title":"feat(kvcache): add nvfp4 kv quantization","updated_at":"2026-09-08T00:00:00Z","url":"https://github.com/FlashML-org/FreeToken/pull/408"},"FlashML-org/FreeToken#409":{"body_hash":"ae748f8c8ddac88b","comments":1,"kind":"issue","number":409,"repo":"FlashML-org/FreeToken","state":"open","title":"Qwen3.8-Flash-Next-NVFP4 cannot start on a 12 GB RTX 3060: ~9.9 GiB of unquantized (BF16) attention/embedding weights leave no room for MoE cache + KV","updated_at":"2026-09-07T11:40:42Z","url":"https://github.com/FlashML-org/FreeToken/issues/409"},"FlashML-org/FreeToken#410":{"body_hash":"6844e6a7d78e03a2","comments":2,"kind":"issue","number":410,"repo":"FlashML-org/FreeToken","state":"open","title":"[Model request] KAT-Coder-V2.5-Dev NVFP4 community exports (compressed-tensors MoE): dense-branch misroute + GDN granularity","updated_at":"2026-09-08T09:15:23Z","url":"https://github.com/FlashML-org/FreeToken/issues/410"},"FlashML-org/FreeToken#413":{"body_hash":"f5fc80d163252223","comments":0,"kind":"pr","number":413,"repo":"FlashML-org/FreeToken","state":"open","title":"fix(models): support llm-compressor NVFP4 MoE export variants","updated_at":"2026-09-08T03:23:36Z","url":"https://github.com/FlashML-org/FreeToken/pull/413"},"FlashML-org/FreeToken#50":{"body_hash":"600c9fb99d6b206f","comments":1,"kind":"issue","number":50,"repo":"FlashML-org/FreeToken","state":"open","title":"Windows: server fails during CUDA graph warmup because dll is not found","updated_at":"2026-09-04T09:19:10Z","url":"https://github.com/FlashML-org/FreeToken/issues/50"},"FlashML-org/FreeToken#60":{"body_hash":"9ff8ed296bc092d9","comments":20,"kind":"issue","number":60,"repo":"FlashML-org/FreeToken","state":"open","title":"Feature request: AMD GPU support","updated_at":"2026-09-07T15:16:21Z","url":"https://github.com/FlashML-org/FreeToken/issues/60"},"FlashML-org/FreeToken#62":{"body_hash":"6cc245487090a7a1","comments":4,"kind":"issue","number":62,"repo":"FlashML-org/FreeToken","state":"open","title":"Offloaded MoE ignores tensor parallelism: TP=2 is a regression, and the fix is not expert parallelism","updated_at":"2026-09-05T23:27:45Z","url":"https://github.com/FlashML-org/FreeToken/issues/62"},"FlashML-org/FreeToken#68":{"body_hash":"f0cf473f9e35a194","comments":3,"kind":"issue","number":68,"repo":"FlashML-org/FreeToken","state":"closed","title":"[Bug] 原始权重转换 FTW 格式卡在"准备权重"——DeepSeek-V4-Flash FP4 超 10 小时无进展","updated_at":"2026-09-04T09:09:39Z","url":"https://github.com/FlashML-org/FreeToken/issues/68"},"FlashML-org/FreeToken#97":{"body_hash":"1d5f83de197791fd","comments":9,"kind":"issue","number":97,"repo":"FlashML-org/FreeToken","state":"open","title":"Difficult to load model despite more than enough vram/ram available","updated_at":"2026-09-08T09:11:21Z","url":"https://github.com/FlashML-org/FreeToken/issues/97"}},"errors":["dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID E006:7D145:25683C4:7A03EB6:6A9FF005 and timestamp 2026-09-08 11:22:45 UTC. For more on scraping GitHub and how it may affect your rights, please review our Terms of Service (https://docs.github.com/en/site-policy/github-terms/github-terms-of-service)"],"external_tracked_issues":{"ROCm/legacy-rocm-build#6461":{"body_hash":"0ac6e80ee9b489cd","comments":4,"kind":"issue","number":6461,"relevance":"high","relevance_reason":"Windows + gfx1201 + hipBLASLt/rocBLAS sequence/state-dependent GEMM failure","repo":"ROCm/legacy-rocm-build","state":"open","title":"[Issue]: rocBLAS bf16/f16 GEMM fails with hipErrorInvalidValue on 2nd call, gfx1201 / Windows / ROCm 7.14","updated_at":"2026-08-25T10:25:55Z","url":"https://github.com/ROCm/legacy-rocm-build/issues/6461"}},"last_checked":"2026-09-08T11:22:19+00:00","releases":{"FlashML-org/FreeToken":{"draft":false,"name":"v0.1.2","prerelease":false,"published_at":"2026-08-19T06:20:18Z","tag":"v0.1.2","url":"https://github.com/FlashML-org/FreeToken/releases/tag/v0.1.2"},"ROCm/ROCm":{"draft":false,"name":"ROCm 7.14.0 Release","prerelease":false,"published_at":"2026-07-16T03:28:58Z","tag":"rocm-7.14.0","url":"https://github.com/ROCm/legacy-rocm-build/releases/tag/rocm-7.14.0"},"ROCm/TheRock":{"draft":false,"name":"ROCm 10.0.0 Release","prerelease":false,"published_at":"2026-08-26T10:03:34Z","tag":"therock-10.0","url":"https://github.com/ROCm/TheRock/releases/tag/therock-10.0"},"apache/tvm-ffi":{"draft":false,"name":"v0.1.13-post3","prerelease":false,"published_at":"2026-08-10T13:31:09Z","tag":"v0.1.13-post3","url":"https://github.com/apache/tvm-ffi/releases/tag/v0.1.13-post3"},"ggml-org/llama.cpp":{"draft":false,"name":"v0.4.0","prerelease":false,"published_at":"2026-09-04T19:56:47Z","tag":"v0.4.0","url":"https://github.com/ggml-org/llama.cpp/releases/tag/v0.4.0"},"triton-lang/triton-windows":{"draft":false,"name":"v3.8.0-windows.post28","prerelease":false,"published_at":"2026-08-29T12:24:14Z","tag":"v3.8.0-windows.post28","url":"https://github.com/triton-lang/triton-windows/releases/tag/v3.8.0-windows.post28"}},"tracked_issues":{"110":{"body_hash":"d6672f16a90f2854","comments":2,"kind":"issue","number":110,"repo":"FlashML-org/FreeToken","state":"open","title":"Unhandled CUDA OOM in prefill expert workspace kills the engine; server keeps accepting requests that never return (root cause behind FlashML-org#20-style headless state; likely also FlashML-org#72)","updated_at":"2026-09-05T23:28:27Z","url":"https://github.com/FlashML-org/FreeToken/issues/110"},"111":{"body_hash":"b01b977d143435c7","comments":3,"kind":"issue","number":111,"repo":"FlashML-org/FreeToken","state":"open","title":"Requests longer than the KV pool are queued forever with no error — and --moe-cache-auto leaves only ~8k tokens of KV on a 96GB GPU","updated_at":"2026-08-31T02:09:35Z","url":"https://github.com/FlashML-org/FreeToken/issues/111"},"120":{"body_hash":"a7caa8c0bca18470","comments":1,"kind":"issue","number":120,"repo":"FlashML-org/FreeToken","state":"open","title":"Native Windows is classified as pin-uncapped, so FlashML-org#112's per-layer residency never auto-engages","updated_at":"2026-08-24T03:04:21Z","url":"https://github.com/FlashML-org/FreeToken/issues/120"},"122":{"body_hash":"4d42515b782747ed","comments":6,"kind":"issue","number":122,"repo":"FlashML-org/FreeToken","state":"open","title":"Windows/ROCm on RX 9060 XT (gfx1200): packed GGUF working end-to-end (dense + MoE), root cause of the RDNA4 offload-decode TDR, benchmarks vs llama.cpp","updated_at":"2026-08-27T07:18:05Z","url":"https://github.com/FlashML-org/FreeToken/issues/122"},"123":{"body_hash":"09dd1b9b8cf6560f","comments":1,"kind":"issue","number":123,"repo":"FlashML-org/FreeToken","state":"closed","title":"Request hang (silent, zero-log) with hybrid_radix cache + moe-backend offload + nvfp4 triton backend — reproduced on 2 independent models, not explained by FlashML-org#110","updated_at":"2026-08-29T05:13:10Z","url":"https://github.com/FlashML-org/FreeToken/issues/123"},"124":{"body_hash":"64a1de8500f135a3","comments":0,"kind":"issue","number":124,"repo":"FlashML-org/FreeToken","state":"closed","title":"NVFP4 checkpoints fail to load: model.safetensors.index.json is not downloaded, but the NVFP4 bank loader requires it","updated_at":"2026-09-02T02:43:20Z","url":"https://github.com/FlashML-org/FreeToken/issues/124"},"79":{"body_hash":"2fe32010327776c6","comments":16,"kind":"issue","number":79,"repo":"FlashML-org/FreeToken","state":"open","title":"FreeToken Roadmap (2026)","updated_at":"2026-09-04T17:33:03Z","url":"https://github.com/FlashML-org/FreeToken/issues/79"},"82":{"body_hash":"ff59a7ff00722b01","comments":17,"kind":"issue","number":82,"repo":"FlashML-org/FreeToken","state":"open","title":"ROCm RDNA4 gfx1201 Ported. Windows only for now. Test please","updated_at":"2026-09-08T10:34:41Z","url":"https://github.com/FlashML-org/FreeToken/issues/82"}},"tracked_prs":{"112":{"base_sha":"f0abe587a11cca53bb3c37a9596fad24973ace62","body_hash":"aa539235301fea3d","checks":{"failed":[],"pending":[],"state":"pending","total":0},"comments":1,"commits":3,"draft":false,"head_sha":"831d38a66bc45543411078c6805de8c42e7603fa","merged":true,"number":112,"review_comments":0,"state":"closed","title":"feat(moe): per-layer host-bank residency","updated_at":"2026-08-24T00:39:32Z","url":"https://github.com/FlashML-org/FreeToken/pull/112"},"118":{"base_sha":"e0a3bbc04652ec0622d932adcbf2772848e3c2e2","body_hash":"d73a76c00cdb2f94","checks":{"failed":[],"pending":[],"state":"pending","total":0},"comments":0,"commits":1,"draft":false,"head_sha":"6241178a7550e5a0ffadd85f0eb04af0160f9988","merged":false,"number":118,"review_comments":0,"state":"open","title":"fix(scheduler): clamp admission to the actual KV pool so oversized prompts fail loudly","updated_at":"2026-08-24T00:52:33Z","url":"https://github.com/FlashML-org/FreeToken/pull/118"},"125":{"base_sha":"bd372b630a028e3faa51f4ab0ef6a98c2f2de501","body_hash":"793491958f083651","checks":{"failed":[],"pending":[],"state":"pending","total":0},"comments":0,"commits":1,"draft":false,"head_sha":"cb44bdeedb7ea6ed2b37c5c49864ad40065e1985","merged":false,"number":125,"review_comments":0,"state":"open","title":"test(pinned): use mapped memory for UVA pointer check","updated_at":"2026-08-24T05:48:44Z","url":"https://github.com/FlashML-org/FreeToken/pull/125"},"129":{"base_sha":"bd372b630a028e3faa51f4ab0ef6a98c2f2de501","body_hash":"d5c4ac29d87557fc","checks":{"failed":[],"pending":[],"state":"pending","total":0},"comments":3,"commits":1,"draft":false,"head_sha":"d30726e006b2708d4fbfead120985bc23c935541","merged":false,"number":129,"review_comments":0,"state":"closed","title":"fix(download): include json files when fetching weights, so NVFP4's index.json is no longer silently skipped","updated_at":"2026-09-02T02:45:39Z","url":"https://github.com/FlashML-org/FreeToken/pull/129"},"131":{"base_sha":"bd372b630a028e3faa51f4ab0ef6a98c2f2de501","body_hash":"33bc52d171e8ca00","checks":{"failed":[],"pending":[],"state":"pending","total":0},"comments":8,"commits":36,"draft":false,"head_sha":"bb432e8c473c557cdef6517abccf0c129bd8c544","merged":false,"number":131,"review_comments":0,"state":"open","title":"GGUF: support all quant types, add qwen35moe","updated_at":"2026-08-29T22:27:37Z","url":"https://github.com/FlashML-org/FreeToken/pull/131"},"132":{"base_sha":"bd372b630a028e3faa51f4ab0ef6a98c2f2de501","body_hash":"f5b92547538a8571","checks":{"failed":[],"pending":[],"state":"pending","total":0},"comments":6,"commits":9,"draft":false,"head_sha":"c0713e44eace3cacd02f27f13d0f6032f4bc86ae","merged":false,"number":132,"review_comments":0,"state":"open","title":"feat(rocm): add RDNA3 and RDNA4 runtime foundation","updated_at":"2026-09-02T10:48:26Z","url":"https://github.com/FlashML-org/FreeToken/pull/132"},"133":{"base_sha":"3a20a79038338c33bd051c52152e6d1faa4d9791","body_hash":"eb17018abe601f2c","checks":{"failed":[],"pending":[],"state":"pending","total":0},"comments":0,"commits":11,"draft":true,"head_sha":"07a352a0ef5c37398f13ef893648dbbfd126f601","merged":false,"number":133,"review_comments":0,"state":"open","title":"fix(rocm): make TVM-FFI index and store JIT kernels portable to HIP","updated_at":"2026-08-31T08:13:20Z","url":"https://github.com/FlashML-org/FreeToken/pull/133"},"134":{"base_sha":"3a20a79038338c33bd051c52152e6d1faa4d9791","body_hash":"467c9a375cece285","checks":{"failed":[],"pending":[],"state":"pending","total":0},"comments":0,"commits":11,"draft":true,"head_sha":"32dfb7bd123d935e256508fbc8c76bbaa9012226","merged":false,"number":134,"review_comments":0,"state":"open","title":"fix(rocm): gate CUDA-only optional backends on ROCm","updated_at":"2026-08-31T08:13:20Z","url":"https://github.com/FlashML-org/FreeToken/pull/134"},"135":{"base_sha":"3a20a79038338c33bd051c52152e6d1faa4d9791","body_hash":"3a0be6c359e61828","checks":{"failed":[],"pending":[],"state":"pending","total":0},"comments":3,"commits":12,"draft":true,"head_sha":"f75a553002d1f78dda2b0a754b55b7676f8a33a2","merged":false,"number":135,"review_comments":0,"state":"open","title":"fix(rocm): route tensor parallel communication through RCCL","updated_at":"2026-09-03T17:04:32Z","url":"https://github.com/FlashML-org/FreeToken/pull/135"},"136":{"base_sha":"3a20a79038338c33bd051c52152e6d1faa4d9791","body_hash":"c875bbbd1cb60d21","checks":{"failed":[],"pending":[],"state":"pending","total":0},"comments":2,"commits":13,"draft":true,"head_sha":"7b3acf2f3d3d321ba7127d65168f9882d2a5d76f","merged":false,"number":136,"review_comments":0,"state":"open","title":"feat(rocm): enable native GGUF kernels on ROCm (validated on RDNA4)","updated_at":"2026-09-02T18:14:05Z","url":"https://github.com/FlashML-org/FreeToken/pull/136"},"137":{"base_sha":"f7c31e92dbf296de3a5bb1b9c5fcb58f988e3a07","body_hash":"a0597a23e4607628","checks":{"failed":[],"pending":[],"state":"pending","total":0},"comments":0,"commits":1,"draft":false,"head_sha":"5eec2582ce89376e3f5016d47ff73ca1b4e8d445","merged":false,"number":137,"review_comments":0,"state":"open","title":"feat(rocm): serve on AMD GPUs through the HIP toolchain","updated_at":"2026-08-26T09:03:28Z","url":"https://github.com/FlashML-org/FreeToken/pull/137"},"27":{"base_sha":"0ab982f10905fa775962a4eddcb44caa50065251","body_hash":"e85637efb5176c06","checks":{"failed":[],"pending":[],"state":"pending","total":0},"comments":4,"commits":1,"draft":false,"head_sha":"d76d5e0105235e9f13edd534fec3dfa86f25cb15","merged":false,"number":27,"review_comments":0,"state":"closed","title":"feat(moe): partial-pin serving to beat the Windows/WSL2 pinned-memory quota (--pin-exempt-layers)","updated_at":"2026-08-24T01:28:09Z","url":"https://github.com/FlashML-org/FreeToken/pull/27"}},"version":1}

Technical baseline updated automatically: 2026-09-08T11:22:19+00:00. This comment is edited in place and does not represent a notification.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-24T15:02:31+00:00).

• [Issue] Windows gfx120X-all Test Artifacts: hipInfo.exe hangs / GPU not detected → cascade failure across 27 jobs: обновлён (state open→open, comments 5→6).
• Новый релевантный issue: bench_offload_cache_copy.py default sweep aborts on minimax-m2.5-marlin and skips the last four profiles.
• Новый релевантный issue: [XPU] inductor workloads can fail intermittently with UR_RESULT_ERROR_PROFILING_INFO_NOT_AVAILABLE (error 17).
• [BUG] preparing weights when converting: обновлён (state open→open, comments 3→4).

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-24T16:56:53+00:00).

• fix(download): include json files when fetching weights, so NVFP4's index.json is no longer silently skipped: обновлены описание/review/discussion (comments 0→1, review 0→0).
• Новый релевантный issue: AssertionError: cache budget too small.
• torch.nn.LSTM.forward produces incorrect outputs (or crashes) on ROCm: обновлён (state open→open, comments 13→14).
• bench_offload_cache_copy.py default sweep aborts on minimax-m2.5-marlin and skips the last four profiles: обновлён (state open→open, comments 0→1).
• feat(server): Add OpenAI /v1/models/{model_id} endpoint: обновлён (state open→open, comments 1→2).
• [Issue]: Windows: rocm-sdk silently drops device files from the devel tree when the link target exceeds MAX_PATH: обновлён (state open→closed, comments 0→0).

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-24T18:31:06+00:00).

• fix(download): include json files when fetching weights, so NVFP4's index.json is no longer silently skipped: обновлены описание/review/discussion (comments 1→2, review 0→0).
• Новый релевантный pr: fix(bench): skip an unsupported slot count instead of aborting the sweep.
• Новый релевантный issue: Eval bug: Windows HIP gfx1201: Qwen3.6-35B-A3B NVFP4 loads, then first MUL_MAT fails with ROCm invalid argument.
• bench_offload_cache_copy.py default sweep aborts on minimax-m2.5-marlin and skips the last four profiles: обновлён (state open→open, comments 1→5).
• Новый релевантный pr: fix(kernel): clear the latched CUDA error before raising in pinned_tensor.
• Новый релевантный pr: CUDA: size routed MoE MMQ N-tiles from typical expert width on RDNA3.
• Новый релевантный issue: Eval bug: inline </think> tag contained in thinking block gets mis-parsed as the end of reasoning content.
• Новый релевантный issue: Misc. bug: RPC + Metal: main node retains the full model with default mmap load mode, capping split models to a single node's memory.
• Новый релевантный pr: AVX2: Speed up large batch size prompt processing of IQ models.
• Новый релевантный pr: CUDA: extend MOE fusion to specdec, earlier MOE glu fusion and topk-router fusion were restricted to 1 token.
• Новый релевантный issue: Feature Request: server in Router Mode: Multiple presets for the same model (loaded once).
• Новый релевантный issue: SYCL: MoE expert tensors never get the Q8_0 reorder path (opt_for_reorder_id).
• Новый релевантный issue: HIP/ROCm backend produces corrupted output on gfx1151: two dense architectures fail, Vulkan is correct with byte-identical flags.
• Новый релевантный pr: CUDA: XOR swizzle flash attn K,V smem fp16 tiles.
• Новый релевантный pr: CUDA: Update Q4_K and Q5_K to use branchless computation, which stops the scale unpack being re-executed for every column in mmvq, improving perf at batch sizes > 1.
• Новый релевантный pr: vulkan: add f16 B-type matmul pipelines and warp tile size tuning for Intel coopmat1.
• Новый релевантный issue: Vulkan: batched decode throughput cliff at n_tokens=9 on many-expert MoE (fixed 8-token thresholds in MMV dispatch).
• Новый релевантный issue: Eval bug: Unexpected massive performace drop with Vulkan Flash Attention.
• Новый релевантный pr: mamba2 : Flatten in/out projections to dispatch GEMM instead of GEMV.
• Новый релевантный issue: Perf: CUDA: Mamba2 batched decode dispatch GEMV instead of GEMM.
• Новый релевантный pr: hexagon: add alternative Hexagon NPU backend implementation.
• Новый релевантный pr: tests: add mul_mat, flash-attention and copy shapes that real models emit.
• Новый релевантный pr: CUDA: Add experimental SM120 CUTLASS MoE prefill for MXFP4 and NVFP4.
• Новый релевантный issue: Misc. bug: Q6_K MUL_MAT produces wrong results on Adreno 740 OpenCL - test-backend-ops fails.
• Новый релевантный issue: Documented Linux HIP Compile Command Fails with pip-installed ROCm.
• Новый релевантный pr: opencl: use right-sized matmul kernels for speculative decoding's verification step.
• Новый релевантный issue: [CUDA][Blackwell] Qwen3.8-27B-NVFP4 decode hangs, CPU spin, no GPU work.
• Новый релевантный pr: metal : skinny-batch mul_mm tiles + deterministic split-K for small-batch decode.
• Новый релевантный issue: Vulkan dense prefill on gfx1151 (RDNA3 iGPU): ~1.8x from a larger K-quant int-mmq tile.
• Новый релевантный issue: Eval bug: MMVQ host/device nwarps mismatch causes NaN on gfx11-generic / RDNA3.5.
• Новый релевантный pr: opencl: tune the quant paths for Intel Xe-LP GPUs to improve its TG and PP performance.
• Новый релевантный pr: opencl: build one program per kernel where the compiler cannot hold several (for Adreno A6x).
• Новый релевантный pr: convert: fix get block count error for Nemotron 3 Ultra.
• Новый релевантный pr: Quant: OCP FP8 E4M3 support.
• Новый релевантный pr: ggml: add support for MXFP8 CPU.
• …ещё 11 изменений скрыто, чтобы комментарий не превращался в свалку.

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-24T19:01:35+00:00).

• CUDA: size routed MoE MMQ N-tiles from typical expert width on RDNA3: обновлён (state open→open, comments 12→14).
• CUDA: XOR swizzle flash attn K,V smem fp16 tiles: обновлён (state open→open, comments 21→22).
• [BUG] preparing weights when converting: обновлён (state open→open, comments 4→5).

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-24T20:49:53+00:00).

• GGUF: support all quant types, add qwen35moe: новый commit 0f26f7ec → 6f6c8640.
• FreeToken Roadmap (2026): comments 2→3.
• Windows/ROCm on RX 9060 XT (gfx1200): packed GGUF working end-to-end (dense + MoE), root cause of the RDNA4 offload-decode TDR, benchmarks vs llama.cpp: comments 0→1.
• Новый релевантный pr: [ROCm][Inductor] Add gfx950 FlexAttention backward default configs.
• Новый релевантный pr: quantization: complete MXFP4 and NVFP4.
• feat(ci): support runtime-only host-ASAN presubmit builds and tests: обновлён (state open→open, comments 4→5).
• Новый релевантный issue: [RFC] Transition ROCm nightly wheels to wheel-based ROCm install (7.14+).
• CUDA: extend MOE fusion to specdec, earlier MOE glu fusion and topk-router fusion were restricted to 1 token: обновлён (state open→open, comments 2→3).
• Новый релевантный pr: Fix hipify skipping sources under Windows subst drives/symlinks in CUDAExtension.

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-24T22:45:58+00:00).

• Новый релевантный pr: [ROCm][c10d] Enable NCCL2 windows and enforce allocator provenance.
• Новый релевантный pr: [ROCm][Inductor] Apply tt.pointer_range guards to template kernels.
• [ROCm][Inductor] Add gfx950 FlexAttention backward default configs: обновлён (state open→open, comments 6→8).
• Новый релевантный issue: Request: RX 6000 (RDNA2) Windows support - native runtime proven (gfx1030/1031/1032).
• Новый релевантный issue: DISABLED test_workspace_runtime_integration (main.TestNVUniversalGemmEpilogueFusion).
• Новый релевантный issue: Please help support older hardware (1080/2080 series).

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-25T01:53:38+00:00).

• FreeToken Roadmap (2026): comments 3→4.
• [ROCm] Fix hipify skipping sources under Windows subst drives/symlinks in CUDAExtension: обновлён (state open→open, comments 3→3).
• [ROCm][c10d] Enable NCCL2 windows and enforce allocator provenance: обновлён (state open→open, comments 3→3).
• Новый релевантный issue: Eval bug: Vulkan regression on AMD Strix Halo UMA between b9660 and b9672: model loads on GPU but inference shows no GPU compute.
• Новый релевантный issue: Eval bug: CUDA error: unsupported value or parameter in cublasSgemm_v2 during large context processing.
• Новый релевантный issue: Eval bug: Starting from B9992, long contexts trigger a CUDA error: the function failed to launch on the GPU.
• feat(server)!: add --gpu to choose the GPU on multi-GPU machines: обновлён (state open→closed, comments 0→1).
• Difficult to load model despite more than enough vram/ram available: обновлён (state open→open, comments 5→6).
• FlashML-org/FreeToken:main: bd372b63 → 2757bb5f — feat(server)!: add --gpu to choose the GPU on multi-GPU machines (feat(server)!: add --gpu to choose the GPU on multi-GPU machines FlashML-org/FreeToken#117).
• Upstream main: релевантные изменения bd372b63 → 2757bb5f — files: python/freetoken/engine/engine.py, python/freetoken/kernel/triton/e4m3_compat.py, python/freetoken/kernel/triton/sampling.py, python/freetoken/moe/bench_profile.py, python/freetoken/moe/benchbw.py, python/freetoken/moe/host_banks.py, python/freetoken/scheduler/scheduler.py.

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-25T03:13:46+00:00).

• FreeToken Roadmap (2026): comments 4→5.
• Новый релевантный issue: GGUF loader only supports gemma4 — Qwen2 / Qwen2.5 GGUF fails with "architecture 'qwen2' is not supported".
• CUDA: XOR swizzle flash attn K,V smem fp16 tiles: обновлён (state open→open, comments 22→23).
• Новый релевантный issue: Eval bug: Excessive compute buffer reservation in MTP draft context on ROCm HIP unnecessarily reduces fitted context size.
• Prefill hangs on RTX 4060 Ti (SM89) for prompts beyond warmup length: обновлён (state open→open, comments 2→3).

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-25T04:55:02+00:00).

• GGUF: support all quant types, add qwen35moe: новый commit 6f6c8640 → b2f84751.
• Новый релевантный issue: Add explicit detection/support for compressed-tensors NVFP4A16 DeepSeek-V4 REAM checkpoints.
• Новый релевантный issue: Qwen3.5 MoE modelopt_mixed loader assumes shared experts use routed-expert NVFP4 quantization.
• [Bug]: Detect hardware fails with cudaErrorNotSupported on NVIDIA RTX 4060 Lapt: обновлён (state open→closed, comments 2→2).
• GGUF loader only supports gemma4 — Qwen2 / Qwen2.5 GGUF fails with "architecture 'qwen2' is not supported": обновлён (state open→closed, comments 0→1).
• CUDA: extend MOE fusion to specdec, earlier MOE glu fusion and topk-router fusion were restricted to 1 token: обновлён (state open→open, comments 3→4).
• Новый релевантный issue: Eval bug: MTP retains inter-request state causing non-deterministic output and model degradation (Qwen3.6-35B-A3B-MTP).
• Новый релевантный issue: Arbitrary Code Execution in PyTorch via torch.package.PackageImporter.load_pickle.
• Новый релевантный issue: torch.distributions.kl_divergence Fails with MultivariateNormal in Dynamo Due to _infer_size Type Error.
• Новый релевантный issue: Model catalog shows models the local machine cannot run (no capability pre-check before display).

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-25T07:09:50+00:00).

• Новый релевантный pr: ggml-cuda: enable MMA FlashAttention for head dim 256 on AMD RDNA.
• Новый релевантный pr: CI: fix workflow_dispatch runner fallback for PyTorch wheel _ci builds + tolerate newer ninja.
• [ROCm][inductor] Fix Inductor compiler selection for ROCm on Windows: обновлён (state open→open, comments 5→7).
• Новый релевантный pr: HIP: RDNA 4 MUL_MAT optimizations.
• Новый релевантный pr: CUDA: enable the CUB path on HIP via hipCUB.
• Новый релевантный issue: Feature request: AMD GPU support.
• Новый релевантный issue: Vulkan: model load hangs indefinitely on RDNA4 (gfx1201) with plain Unsloth Q6_K quant, or OOMs host with --direct-io.
• Новый релевантный pr: cuda : dynamic MMVQ nwarps for narrow matrices.

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-25T09:01:32+00:00).

• GGUF: support all quant types, add qwen35moe: новый commit b2f84751 → dd48aacb.
• Новый релевантный pr: Make the drafter probabilistic and the target verify by rejection sampling for simple draft and MTP.
• Новый релевантный issue: Eval bug: qwen35moe (hybrid recurrent/attention) models produce empty generation when prompt exceeds ~16K tokens via llama-server (Metal).
• Новый релевантный pr: fix(engine): load prefill triton kernels at startup, not mid-request.
• Новый релевантный issue: CUDA OoM with larger --kv-reserve-tokens (workaround: empiric lower --memory-ratio).
• Новый релевантный pr: Vulkan: add hoisting support for row IDs and expert count in shaders.
• Новый релевантный pr: server: add recurrent state shrink/expand for prompt cache (#22746).
• Новый релевантный pr: vulkan : transpose A-matrix data layout for K-quant mul_mat performance.
• Новый релевантный issue: Silent GTT fallback when context + MTP exceeds VRAM — no error at load, massive throughput collapse at runtime.
• Новый релевантный issue: Eval bug: --batch-size and --ubatch-size ignored when set to 1024/1024.
• Новый релевантный issue: draft-mtp speculative decoding not activating on Turing (sm_75) with hybrid SSM+attention model (Qwen3.6-35B-A3B).
• Новый релевантный pr: WIP: kv-cache apply_ubtach change from O(n) to O(1).
• Новый релевантный issue: Eval bug: vulkan performance drop pp512 560 -> 180.
• Новый релевантный pr: llama + spec: MTP Support.
• CUDA: enable the CUB path on HIP via hipCUB: обновлён (state open→open, comments 46→47).

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-25T10:51:05+00:00).

• HIP/ROCm backend produces corrupted output on gfx1151: two dense architectures fail, Vulkan is correct with byte-identical flags: обновлён (state open→open, comments 7→8).
• Eval bug: Windows HIP gfx1201: Qwen3.6-35B-A3B NVFP4 loads, then first MUL_MAT fails with ROCm invalid argument: обновлён (state open→open, comments 0→4).
• Новый релевантный issue: Long prefill (>48k tokens) crashes the whole server — CUDA driver error: device not ready in fla/l2norm kernel.
• Новый релевантный issue: [Issue] Linux gfx103X: GPU not accessible to runner — rocminfo exits 8 / "No GPU found matching.
• Новый релевантный issue: torch.cholesky_inverse fails with HIPBLAS_STATUS_ALLOC_FAILED at batch 20 on ROCm due to fixed hipBLAS workspace.
• Новый релевантный pr: [ROCm] Enable CK SDPA on gfx1200/gfx1201 (RDNA4).
• Новый релевантный issue: [Issue]: Access violation (0xC0000005) in bmm/baddbmm on gfx1201 (RX 9070 XT) under Windows, crashes in both hipBLASLt and rocBLAS.
• Новый релиз: ggml-org/llama.cpp v0.3.0.

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-25T11:11:49+00:00).

• Новый явно отслеживаемый external issue: [Issue]: rocBLAS bf16/f16 GEMM fails with hipErrorInvalidValue on 2nd call, gfx1201 / Windows / ROCm 7.14. High relevance.
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 8024:3880E4:B12B:25494:6A8D788F and timestamp 2026-08-25 11:12:15 UTC. For…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-25T13:13:08+00:00).

• feat(rocm): add RDNA3 and RDNA4 runtime foundation: обновлены описание/review/discussion (comments 1→2, review 0→0).
• Новый релевантный issue: Feature: Predictive Expert Offloading.
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID F440:3A3DE2:D8F1C6:2D7E43E:6A8D950E and timestamp 2026-08-25 13:13:50 UTC.…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-25T15:07:11+00:00).

• FreeToken Roadmap (2026): comments 5→6.
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 0407:272BC6:1AC4A9F:5964A5E:6A8DAFBB and timestamp 2026-08-25 15:07:39 UTC…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-25T16:57:34+00:00).

• Model catalog shows models the local machine cannot run (no capability pre-check before display): обновлён (state open→open, comments 0→1).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 283F:2B3BAA:2009A4A:6A360EB:6A8DC9A1 and timestamp 2026-08-25 16:58:09 UTC…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-25T18:58:44+00:00).

• Новый релевантный pr: fix(deps): use PyTorch 2.13 to fix SM89 hangs.
• Новый релевантный issue: Apodex-1.1-mini-NVFP4 fails: dense_quant hardcoded to NVFP4 when experts are NVFP4, but shared_expert is FP8 (modelopt MIXED_PRECISION).
• Difficult to load model despite more than enough vram/ram available: обновлён (state open→open, comments 6→7).
• Новый релевантный issue: FP8 row-wise _scaled_mm hangs on SM89 with PyTorch 2.11.
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID C819:1EE7B3:2EC3E0B:9D340E1:6A8DE60D and timestamp 2026-08-25 18:59:25 UTC…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-25T20:47:37+00:00).

• fix(deps): use PyTorch 2.13 to fix SM89 hangs: обновлён (state open→open, comments 0→0).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 7830:16ACDE:3C41178:CA29304:6A8DFF87 and timestamp 2026-08-25 20:48:07 UTC…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-25T22:46:42+00:00).

• GGUF: support all quant types, add qwen35moe: новый commit dd48aacb → 9952a39d.
• Maxritz/FreeToken-rocm-test:main: 45675e47 → 3a5616ec — Merge pull request FreeToken ROCm Watch — notification channel #1 from Castoff995/codex/gfx1201-offload-decode.
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID CC18:2EB04F:4D5504F:100358CF:6A8E1B72 and timestamp 2026-08-25 22:47:14 UT…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-26T01:59:36+00:00).

• GGUF: support all quant types, add qwen35moe: новый commit 9952a39d → 7da6b77e.
• feat(rocm): enable native GGUF kernels on RDNA4: обновлены описание/review/discussion (comments 0→1, review 0→0).
• Requests longer than the KV pool are queued forever with no error — and --moe-cache-auto leaves only ~8k tokens of KV on a 96GB GPU: comments 1→2.
• Новый релевантный pr: feat(kvcache): 8-bit KV cache (q8_0 / fp8_e4m3) behind --kv-cache-dtype.
• FlashML-org/FreeToken:main: 2757bb5f → f7c31e92 — fix(fla): stop l2norm recompiling per token count.
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 7431:947DB:5671D36:1210F83B:6A8E48B1 and timestamp 2026-08-26 02:00:17 UTC…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-26T03:19:34+00:00).

• feat(rocm): add RDNA3 and RDNA4 runtime foundation: обновлены описание/review/discussion (comments 2→3, review 0→0).
• feat(rocm): serve on AMD GPUs through the HIP toolchain: новый commit 7e1d7421 → 5eec2582.
• Новый релевантный issue: Qwen3.6-35B-A3B hits KV-cache limit during prolonged reasoning on LeetCode problem.
• Model catalog shows models the local machine cannot run (no capability pre-check before display): обновлён (state open→open, comments 1→2).
• Новый релевантный issue: [AppImage] Global LD_LIBRARY_PATH export breaks system-curl child processes (engine installer fails on non-Ubuntu distros).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID B00B:2DC8AF:65DCB0D:15086008:6A8E5B63 and timestamp 2026-08-26 03:20:03 UT…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-26T04:57:21+00:00).

• Новый релевантный issue: [Linux] AppImage blank window / WebKitWebProcess abort: "Could not create default EGL display: EGL_BAD_PARAMETER" (bundled libwayland-client on Mesa 25+ hosts).
• Новый релевантный issue: Loading nvidia/Qwen3.6-35B-A3B-NVFP4 fails: "Incomplete fp8 fusions: ['model.layers.37.linear_attn.in_proj_qkvz']".
• Qwen3.6-35B-A3B hits KV-cache limit during prolonged reasoning on LeetCode problem: обновлён (state open→open, comments 0→0).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID D828:1EDC:4C5ACCD:FD5C08A:6A8E724C and timestamp 2026-08-26 04:57:48 UTC. …

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-26T07:10:22+00:00).

• GGUF: support all quant types, add qwen35moe: новый commit 7da6b77e → bb432e8c.
• FreeToken Roadmap (2026): body edited.
• Новый релевантный pr: qwen3_5_moe: load mixed-precision compressed-tensors checkpoints (unsloth NVFP4+FP8).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID CAC4:33DF9:7ABF9D8:19628807:6A8E9187 and timestamp 2026-08-26 07:11:03 UTC…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-26T09:04:06+00:00).

• Bug in PyTorch 2.11 causes SM89 hangs: обновлён (state open→open, comments 0→0).
• feat(models): support TP for qwen3_5_moe: обновлён (state open→open, comments 2→3).
• FlashML-org/FreeToken:main: f7c31e92 → 9ef36513 — chore(assets): update wechat group QR code.
• Maxritz/FreeToken-rocm-test:main: 3a5616ec → 89f14790 — Merge pull request Adding Developer Contributions from Pial Ghosh Maxritz/FreeToken-ROCm#2 from PialGhosh2233/main.
• PialGhosh2233/FreeToken-rocm-gfx1200:gfx1200-moe-gguf: 544b83df → a338f49b — docs: Gemma-4-26B-A4B QAT q4_0 GGUF results on gfx1200.
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 0846:30A2EB:30974E:A35D01:6A8EAC23 and timestamp 2026-08-26 09:04:35 UTC. …

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-26T10:52:44+00:00).

• Новый релевантный pr: feat(models): add Qwen3-Next-80B-A3B support.
• feat(models): support TP for qwen3_5_moe: обновлён (state open→open, comments 3→5).
• Новый релиз: ROCm/TheRock therock-10.0.
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID B008:2BB36A:73A408:77B54F:6A8EC59E and timestamp 2026-08-26 10:53:18 UTC. …

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-26T13:18:13+00:00).

• FreeToken Roadmap (2026): comments 6→7.
• Feature request: AMD GPU support: обновлён (state open→open, comments 11→12).
• feat(models): support TP for qwen3_5_moe: обновлён (state open→open, comments 5→7).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 0400:15153F:A5D123:230714B:6A8EE7C0 and timestamp 2026-08-26 13:18:56 UTC.…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-08-26T15:55:00+00:00).

• [Linux] AppImage blank window / WebKitWebProcess abort: "Could not create default EGL display: EGL_BAD_PARAMETER" (bundled libwayland-client on Mesa 25+ hosts): обновлён (state open→open, comments 0→1).
• feat(models): support TP for qwen3_5_moe: обновлён (state open→open, comments 7→8).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID D82D:361ECC:232D6C:2C6E23:6A8F0C76 and timestamp 2026-08-26 15:55:34 UTC. …

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-02T21:05:16+00:00).

• feat(rocm): enable native GGUF kernels on ROCm (validated on RDNA4): обновлены описание/review/discussion (comments 1→2, review 0→0).
• FreeToken-Intel: 35B MoE, live tokens off an Arc Pro B70 (SYCL / XPU port): обновлён (state open→open, comments 0→1).
• fix: [BUG] qwen3_5 dense: mixed-precision NVFP4 crashes in ct_bf16_fuse (Float8 × BFloat16 promotion): обновлён (state open→open, comments 2→3).
• [BUG] qwen3_5 dense: mixed-precision NVFP4 crashes in ct_bf16_fuse (Float8 × BFloat16 promotion): обновлён (state open→open, comments 2→3).
• ROCm: validate native gfx1151 serving on Strix Halo: обновлён (state open→open, comments 5→5).
• Новый релевантный pr: rocm: fix native extensions, kernel JIT builds, and Triton PTX fallbacks for gfx1150.
• Can't load Qwen3.8-Flash-Next: обновлён (state open→closed, comments 0→1).
• feat(kvcache): sub-byte Q4_0 and Q6_0 KV cache quantization: обновлён (state open→open, comments 5→6).
• Новый релевантный issue: Add support for all major 2026 KV cache compression techniques (TurboQuant, RotorQuant, KIVI, KVQuant, eviction methods).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 5C10:1E435E:27CAD87:8168886:6A988FAF and timestamp 2026-09-02 21:05:51 UTC…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-03T00:18:06+00:00).

• feat(models): support TP for qwen3_5_moe: обновлён (state open→open, comments 12→13).
• feat(moe): NVMe disk tier for MoE expert banks: обновлён (state open→open, comments 3→4).
• Long-context decode: client wall-clock is ~half the scheduler gen throughput: обновлён (state open→open, comments 0→1).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID F40A:379321:63FB4E:7363C1:6A98BCDF and timestamp 2026-09-03 00:18:39 UTC. …

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-03T07:12:00+00:00).

• Новый релевантный issue: [Linux] AppImage blank window / WebKitWebProcess abort: "Could not create default EGL display: EGL_BAD_PARAMETER" (bundled libwayland-client on Mesa 25+ hosts).
• Can't load Qwen3.8-Flash-Next: обновлён (state closed→open, comments 1→4).
• Новый релевантный pr: feat(models): add Qwen3-Next-80B-A3B support.
• FreeToken-Intel: 35B MoE, live tokens off an Arc Pro B70 (SYCL / XPU port): обновлён (state open→open, comments 1→3).
• feat(kvcache): store the KV cache as fp8 e4m3 codes (--kv-cache-dtype…: обновлён (state open→open, comments 1→3).
• feat(moe): NVMe disk tier for MoE expert banks: обновлён (state open→open, comments 4→7).
• Новый релевантный pr: feat(sm70): support Volta / Tesla V100 via torch 2.10 cu128 downgrade.
• Новый релевантный pr: fix(gemma4): support dense Gemma-4 GGUF checkpoints.
• Новый релевантный issue: gemma4 GGUF: dense Gemma-4 checkpoints cannot load — parse_gguf_config requires gemma4.expert_count and hardcodes moe_enabled=True.
• Новый релевантный issue: AssertionError: cache budget too small.
• FlashML-org/FreeToken:main: 6eca2d7d → 03c28d2b — fix(kernel): make Triton top-k/top-p sampling exact (fix(kernel): make Triton top-k/top-p sampling exact FlashML-org/FreeToken#329).
• Upstream main: релевантные изменения 6eca2d7d → 03c28d2b — commits: fix(kernel): make Triton top-k/top-p sampling exact (fix(kernel): make Triton top-k/top-p sampling exact FlashML-org/FreeToken#329) | files: python/freetoken/kernel/triton/sampling.py.
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID DC00:3AD3B5:35AB77:3F9B37:6A991DE3 and timestamp 2026-09-03 07:12:35 UTC. …

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-03T12:50:38+00:00).

• feat(kvcache): store the KV cache as fp8 e4m3 codes (--kv-cache-dtype…: обновлён (state open→open, comments 3→6).
• FreeToken-Intel: 35B MoE, live tokens off an Arc Pro B70 (SYCL / XPU port): обновлён (state open→open, comments 3→4).
• Новый релевантный issue: [Feature Request] Support compressed-tensors per-channel FP8 checkpoints for Qwen3.5 MoE.
• Новый релевантный issue: Qwen3.5-122B-A10B-NVFP4.
• Offloaded MoE ignores tensor parallelism: TP=2 is a regression, and the fix is not expert parallelism: обновлён (state open→open, comments 1→2).
• Новый релевантный issue: TP>1: the first request can be lost in the rank-0→rank-N ZeroMQ PUB/SUB relay if it arrives before rank N's subscription is registered — both ranks then block forever while /v1/models keeps answering 200.
• The two protocol surfaces disagree on image input: /v1/chat/completions 400s, /v1/messages silently drops the image and answers: обновлён (state open→open, comments 0→1).
• Новый релевантный issue: [Feature Request]: Support EXL3 Format.
• Can't load Qwen3.8-Flash-Next: обновлён (state open→closed, comments 4→5).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 1C0C:1CC683:BA780C:254E924:6A996D48 and timestamp 2026-09-03 12:51:20 UTC.…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-03T17:52:58+00:00).

• fix(rocm): route tensor parallel communication through RCCL: обновлены описание/review/discussion (comments 2→3, review 0→0).
• feat(kvcache): store the KV cache as fp8 e4m3 codes (--kv-cache-dtype…: обновлён (state open→open, comments 6→8).
• perf(ple): fuse the n-gram row-id hash into one Triton kernel: обновлён (state open→open, comments 0→1).
• Новый релевантный pr: kernel: clamp zero-size host registrations; WDDM lock-pool hint on pin failure.
• Новый релевантный issue: Unified-memory platforms (DGX Spark, sm_121): moe-backend auto resolves to offload, adding ~130s/request of pure overhead.
• [Feature Request]: Support EXL3 Format: обновлён (state open→open, comments 0→1).
• fix(kernels): drop the D2H sync from the varlen GDN/KDA prefill conv: обновлён (state open→open, comments 0→1).
• feat(moe): NVMe disk tier for MoE expert banks: обновлён (state open→open, comments 7→9).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID BC3E:36AA1F:350DF5F:A928520:6A99B416 and timestamp 2026-09-03 17:53:26 UTC…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-03T21:04:03+00:00).

• Offloaded MoE ignores tensor parallelism: TP=2 is a regression, and the fix is not expert parallelism: обновлён (state open→open, comments 2→3).
• Новый релевантный issue: TP > 1: every rank keeps torch's default intra-op thread count, so a multi-rank weight load busy-waits itself to a standstill on a small-core host (19 min -> 61 s).
• Новый релевантный issue: fix(checkpoint): FTW output drops the PLE n-gram shards, so qwen4_exp checkpoints cannot be served from it.
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID C840:ADD15:1D30A0:219348:6A99E0E5 and timestamp 2026-09-03 21:04:37 UTC. F…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-04T00:07:43+00:00).

• feat(moe): NVMe disk tier for MoE expert banks: обновлён (state open→open, comments 9→10).
• perf(ple): fuse the n-gram row-id hash into one Triton kernel: обновлён (state open→open, comments 1→2).
• fix(kernels): drop the D2H sync from the varlen GDN/KDA prefill conv: обновлён (state open→open, comments 1→2).
• FreeToken-Intel: 35B MoE, live tokens off an Arc Pro B70 (SYCL / XPU port): обновлён (state open→open, comments 4→6).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 403A:238E3:39E9E7:435A5A:6A9A0BF0 and timestamp 2026-09-04 00:08:16 UTC. F…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-04T04:36:59+00:00).

• feat(moe): NVMe disk tier for MoE expert banks: обновлён (state open→open, comments 10→12).
• Новый релевантный pr: feat(bench): pass ft serve options through '--' in bench_decode_moe.
• feat(kvcache): store the KV cache as fp8 e4m3 codes (--kv-cache-dtype…: обновлён (state open→open, comments 8→9).
• fix(kernels): drop the D2H sync from the varlen GDN/KDA prefill conv: обновлён (state open→open, comments 2→3).
• Новый релевантный pr: feat(amd): add Linux HIP/ROCm backend for gfx1201.
• FlashML-org/FreeToken:main: 03c28d2b → 86214a94 — chore: add issue templates with FAQ/Roadmap checks and auto-labels (chore: add issue templates with FAQ/Roadmap checks and auto-labels FlashML-org/FreeToken#375).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID E05B:267108:A19E7:1F803F:6A9A4B05 and timestamp 2026-09-04 04:37:25 UTC. F…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-04T11:21:04+00:00).

• Новый релевантный issue: [Bug]: qwen3_5_moe fails to convert/load compressed-tensors NVFP4 checkpoints with model.language_model prefix.
• Новый релевантный issue: Windows: server fails during CUDA graph warmup because dll is not found.
• Новый релевантный issue: [Bug] 原始权重转换 FTW 格式卡在"准备权重"——DeepSeek-V4-Flash FP4 超 10 小时无进展.
• Новый релевантный issue: NotImplementedError: GPUs on both sides of the sm_89 fp8 boundary in one process: the host-side e4m3 convention is process-global.
• Новый релевантный pr: fix(rocm): make CPU/Hybrid MoE graph replay safe.
• feat(models): support TP for qwen3_5_moe: обновлён (state open→open, comments 13→14).
• feat(kvcache): sub-byte Q4_0 and Q6_0 KV cache quantization: обновлён (state open→open, comments 6→7).
• FlashML-org/FreeToken:main: 86214a94 → af71ba43 — ci: publish engine-.json manifests alongside the nightly wheels (ci: publish engine-<platform>.json manifests alongside the nightly wheels FlashML-org/FreeToken#377).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID B028:11B57D:4FC402:6F321F:6A9AA9C3 and timestamp 2026-09-04 11:21:39 UTC. …

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-04T16:23:27+00:00).

• feat(kvcache): sub-byte Q4_0 and Q6_0 KV cache quantization: обновлён (state open→open, comments 7→10).
• qwen3_5_moe: run lm_head on sampled rows only (fixes 32 GB first-prefill OOM): обновлён (state open→open, comments 0→1).
• feat(kvcache): store the KV cache as fp8 e4m3 codes (--kv-cache-dtype…: обновлён (state open→open, comments 9→11).
• Новый релевантный issue: Build on Windows?.
• Новый релевантный issue: --moe-cache-auto ignores --num-tokens / --num-pages: expert fill overruns the budget and the KV pool OOMs at boot.
• feat(models): support TP for qwen3_5_moe: обновлён (state open→open, comments 14→16).
• fix(engine): reserve the pool's dummy page in the --moe-cache-auto KV floor: обновлён (state open→open, comments 0→1).
• fix(kernels): drop the D2H sync from the varlen GDN/KDA prefill conv: обновлён (state open→open, comments 3→4).
• perf(ple): fuse the n-gram row-id hash into one Triton kernel: обновлён (state open→open, comments 2→3).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID B00D:A411D:19281F4:5043542:6A9AF0A2 and timestamp 2026-09-04 16:24:03 UTC.…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-04T20:46:54+00:00).

• FreeToken Roadmap (2026): comments 15→16.
• --moe-cache-auto ignores --num-tokens / --num-pages: expert fill overruns the budget and the KV pool OOMs at boot: обновлён (state open→open, comments 0→1).
• Новый релевантный pr: feat(qwen4_exp): tensor parallelism for Qwen3.8-Flash-Next (offload backend).
• Новый релевантный issue: Desktop engine installer ignores existing FreeToken engine path and always installs a new venv in AppData.
• Новый релевантный issue: Feature request: AMD GPU support.
• Новый релевантный pr: fix(engine): load prefill triton kernels at startup, not mid-request.
• Новый релевантный pr: fix(scheduler): hybrid radix hit corruption — GDN snapshot copy-on-donate + admission barriers.
• perf(ple): fuse the n-gram row-id hash into one Triton kernel: обновлён (state open→open, comments 3→4).
• Новый релиз: ggml-org/llama.cpp v0.4.0.
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 8D05:8E1C1:BAE087:25AD0BB:6A9B2E59 and timestamp 2026-09-04 20:47:21 UTC. …

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-05T00:08:11+00:00).

• Новый релевантный pr: fix(models): better support for mixed-precision compressed-tensors NVFP4.
• FreeToken-Intel: 35B MoE, live tokens off an Arc Pro B70 (SYCL / XPU port): обновлён (state open→open, comments 6→7).
• Новый релевантный pr: feat(qwen4_exp): load-time per-tensor FP8 dense projections (W8A8 via _scaled_mm).
• Новый релевантный pr: qwen4_exp: load modelopt MIXED_PRECISION (NVFP4 experts + block-FP8 dense) checkpoints.
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID A087:19B230:B271ED:23BB33E:6A9B5D94 and timestamp 2026-09-05 00:08:52 UTC.…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-05T04:33:16+00:00).

• Новый релевантный pr: feat(server): Add OpenAI /v1/models/{model_id} endpoint.
• feat(kvcache): store the KV cache as fp8 e4m3 codes (--kv-cache-dtype…: обновлён (state open→open, comments 11→12).
• Новый релевантный pr: qwen4_exp: serve the block-FP8 dense projections natively (+25% decode).
• feat(qwen4_exp): load-time per-tensor FP8 dense projections (W8A8 via _scaled_mm): обновлён (state open→open, comments 1→1).
• Qwen3.5-122B-A10B-NVFP4: обновлён (state open→closed, comments 0→1).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 5B82:62FA1:6A593:152A34:6A9B9BAC and timestamp 2026-09-05 04:33:48 UTC. Fo…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-05T10:38:07+00:00).

• fix(scheduler): hybrid radix hit corruption — GDN snapshot copy-on-donate + admission barriers: обновлён (state open→closed, comments 2→2).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID FC04:3435C7:32F6CAB:A62AF18:6A9BF137 and timestamp 2026-09-05 10:38:47 UTC…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-05T15:16:33+00:00).

• feat(kvcache): store the KV cache as fp8 e4m3 codes (--kv-cache-dtype…: обновлён (state open→open, comments 12→13).
• Новый релевантный pr: feat(server): OpenAI protocol coverage -- sampling extras, n, echo/suffix, usage details, tokenize/detokenize/metrics.
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID D833:2BFABC:EF5B8C:3082F89:6A9C327A and timestamp 2026-09-05 15:17:14 UTC.…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-05T18:20:15+00:00).

• Новый релевантный issue: [Feature] int4 routed experts: AWQ -> Q4_1 and compressed-tensors pack-quantized -> Q4_0, with no new CUDA kernel (offering a PR).
• Новый релевантный issue: temperature 0 is not reproducible with thinking on (identical requests diverge on near-ties).
• Новый релевантный issue: [Bug] --max-output-tokens is ignored by /v1/chat/completions and /v1/messages (only /v1/responses honours it).
• Новый релевантный issue: Support official nvidia/Qwen3.8-Flash-Next-NVFP4 model.
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 4C29:3AAA5D:3CFA4B:C461CD:6A9C5D87 and timestamp 2026-09-05 18:20:55 UTC. …

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-05T22:04:18+00:00).

• feat(qwen4_exp): load-time per-tensor FP8 dense projections (W8A8 via _scaled_mm): обновлён (state open→open, comments 1→2).
• Новый релевантный pr: feat(kvcache): add ISO3/ISO4 KV-cache quantization (--kv-cache-iso).
• Новый релевантный pr: feat(rocm): AMD ROCm runtime for RDNA3/4 with native Qwen3.5-MoE GGUF decode parity.
• Новый релевантный issue: 5090+256G内存无法运行红帽的GLM5.3FlashNVFP4.
• Новый релевантный pr: Support poolside/Laguna-S-2.1-NVFP4, and fix serving from source on Windows.
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 5C19:9AD18:32FF84:A29644:6A9C9200 and timestamp 2026-09-05 22:04:48 UTC. F…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-05T23:55:10+00:00).

• Unhandled CUDA OOM in prefill expert workspace kills the engine; server keeps accepting requests that never return (root cause behind #20-style headless state; likely also #72): comments 1→2.
• Новый релевантный issue: Feature Request: opt-in FP8 KV-cache support for long-context MoE inference.
• Новый релевантный issue: feat(server): add a server-side thinking override.
• [Bug] --max-output-tokens is ignored by /v1/chat/completions and /v1/messages (only /v1/responses honours it): обновлён (state open→open, comments 0→1).
• TP > 1: every rank keeps torch's default intra-op thread count, so a multi-rank weight load busy-waits itself to a standstill on a small-core host (19 min -> 61 s): обновлён (state open→open, comments 0→1).
• Новый релевантный issue: Long-context decode: client wall-clock is ~half the scheduler gen throughput.
• Offloaded MoE ignores tensor parallelism: TP=2 is a regression, and the fix is not expert parallelism: обновлён (state open→open, comments 3→4).
• Support official nvidia/Qwen3.8-Flash-Next-NVFP4 model: обновлён (state open→open, comments 0→1).
• qwen4_exp: serve the block-FP8 dense projections natively (+25% decode): обновлён (state open→open, comments 1→3).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID A416:8896:164EAC7:4803F70:6A9CABFA and timestamp 2026-09-05 23:55:38 UTC. …

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-06T04:42:36+00:00).

• Новый релевантный issue: --moe-cache-auto leaves no room for prefill: boots and serves decode, then the first 8k prefill OOMs and kills the worker (NVFP4 dense).
• Новый релевантный pr: fix(cpu-moe): honor padded fp8 scale strides, add avx512f tier and float64 parity (builds on #36).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 28C8:F8EC0:3E48B17:C7B2676:6A9CEF59 and timestamp 2026-09-06 04:43:05 UTC.…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-06T11:00:34+00:00).

• [Feature Request] Support compressed-tensors per-channel FP8 checkpoints for Qwen3.5 MoE: обновлён (state open→open, comments 1→2).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 1440:B0E8F:3F8E4:CF3EB:6A9D47FB and timestamp 2026-09-06 11:01:15 UTC. For…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-06T15:29:35+00:00).

• Новый релевантный issue: fix(server): allow overriding the distributed port.
• feat(server): add a server-side thinking override: обновлён (state open→open, comments 2→3).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 980C:1453EF:5ED41CB:13373AF3:6A9D86FC and timestamp 2026-09-06 15:30:04 UT…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-06T18:19:07+00:00).

• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 0445:C9E59:3511EB0:ABF8AC9:6A9DAEBB and timestamp 2026-09-06 18:19:40 UTC.…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-06T22:08:17+00:00).

• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID BBC9:82807:306F530:9C79979:6A9DE469 and timestamp 2026-09-06 22:08:41 UTC.…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-06T23:56:41+00:00).

• feat(kvcache): store the KV cache as fp8 e4m3 codes (--kv-cache-dtype…: обновлён (state open→open, comments 13→14).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 8AAC:3C9966:DF5698:2CA3D46:6A9DFDD9 and timestamp 2026-09-06 23:57:13 UTC.…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-07T04:45:00+00:00).

• Новый релевантный pr: docs(models): record full glm-5.3 checkpoint screening.
• Новый релевантный pr: fix(models): serve qwen4_exp FTW checkpoints with PLE streamed from source.
• Новый релевантный pr: fix(models): support nvidia/Qwen3.8-Flash-Next-NVFP4 in qwen4_exp.
• Новый релевантный issue: prompt is too long: 16535 tokens > 8221 maximum.
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 2847:116CF:6042B:137950:6A9E416D and timestamp 2026-09-07 04:45:33 UTC. Fo…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-07T12:37:24+00:00).

• Новый релевантный pr: feat(kvcache): add nvfp4 kv quantization.
• Новый релевантный issue: Qwen3.8-Flash-Next-NVFP4 cannot start on a 12 GB RTX 3060: ~9.9 GiB of unquantized (BF16) attention/embedding weights leave no room for MoE cache + KV.
• Новый релевантный issue: Qwen3.6‑35B‑A3B‑Uncensored‑HauhauCS‑Aggressive GGUF 加载报错 qwen35moe architecture not supported.
• prompt is too long: 16535 tokens > 8221 maximum: обновлён (state open→closed, comments 0→2).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID B440:96462:3A661F1:BF597F6:6A9EB01E and timestamp 2026-09-07 12:37:50 UTC.…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-07T18:44:43+00:00).

• [Feature Request]: Support EXL3 Format: обновлён (state open→open, comments 1→2).
• Новый релевантный issue: [Model request] KAT-Coder-V2.5-Dev NVFP4 community exports (compressed-tensors MoE): dense-branch misroute + GDN granularity.
• Feature request: AMD GPU support: обновлён (state open→open, comments 19→20).
• Qwen3.5-122B-A10B-NVFP4: обновлён (state closed→closed, comments 1→2).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 448C:19A6E7:197F81C:530B7B1:6A9F0634 and timestamp 2026-09-07 18:45:08 UTC…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-07T22:51:23+00:00).

• docs(models): qualify full GLM-5.3 on H200 Serverless: обновлён (state open→open, comments 0→0).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID B417:14B945:95ACBB:1DD3743:6A9F4013 and timestamp 2026-09-07 22:52:04 UTC.…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-08T04:40:22+00:00).

• Новый релевантный issue: Two blocking bugs in Windows Desktop on consumer Blackwell (RTX 50): FP8 row-wise gate crash + Mistral tokenizer empty Chinese output.
• Новый релевантный issue: [Bug]: RuntimeError: Rowwise scaling is not currently supported on your device.
• Новый релевантный issue: Difficult to load model despite more than enough vram/ram available.
• Новый релевантный issue: FreeToken Engine Install Fails in Windows with RTX4060.
• Новый релевантный issue: [Linux] AppImage blank window / WebKitWebProcess abort: "Could not create default EGL display: EGL_BAD_PARAMETER" (bundled libwayland-client on Mesa 25+ hosts).
• Новый релевантный pr: fix(models): support llm-compressor NVFP4 MoE export variants.
• fix(models): better support for mixed-precision compressed-tensors NVFP4: обновлён (state open→open, comments 0→2).
• [Model request] KAT-Coder-V2.5-Dev NVFP4 community exports (compressed-tensors MoE): dense-branch misroute + GDN granularity: обновлён (state open→open, comments 0→1).
• [Feature Request] Support compressed-tensors per-channel FP8 checkpoints for Qwen3.5 MoE: обновлён (state open→open, comments 2→4).
• Новый релевантный pr: fix: [BUG] qwen3_5 dense: mixed-precision NVFP4 crashes in ct_bf16_fuse (Float8 × BFloat16 promotion).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID 182A:333172:480A450:E6EEF41:6A9F91D0 and timestamp 2026-09-08 04:40:48 UTC…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.

[github-actions]

@Castoff995 FreeToken ROCm Watch: обнаружены существенные изменения (2026-09-08T11:22:19+00:00).

• ROCm RDNA4 gfx1201 Ported. Windows only for now. Test please: comments 16→17.
• [Model request] KAT-Coder-V2.5-Dev NVFP4 community exports (compressed-tensors MoE): dense-branch misroute + GDN granularity: обновлён (state open→open, comments 1→2).
• Difficult to load model despite more than enough vram/ram available: обновлён (state open→open, comments 8→9).
• fix(models): better support for mixed-precision compressed-tensors NVFP4: обновлён (state open→open, comments 2→3).
• ⚠️ Новый сбой части мониторинга: dependency discovery: ApiError: GET https://api.github.com/search/issues?q=repo%3Aggml-org/llama.cpp%20MUL_MAT&sort=updated&order=desc&per_page=20: HTTP 403: API rate limit exceeded for installation. If you reach out to GitHub Support for help, please include the request ID E006:7D145:25683C4:7A03EB6:6A9FF005 and timestamp 2026-09-08 11:22:45 UTC.…

Перед следующим Codex-таском стоит сначала сверить эти изменения с текущим blocker/патчем.