[Mamba][Bugfix] Raise on insufficient cache blocks instead of silently capping cudagraph sizes (#38270)

Nicolò Lucchesi committed 7 months ago

[ROCm] [DOC] Update the Documentation to include ROCm Nightly Wheel support (#38457)

TJian committed 7 months ago

[CI] Fix SPLADE pooler test broken by #38139 (#38495)

haosdent committed 7 months ago

[ROCm] Enable MORI EP for unquantized MoE with AITER backend (#37529)

Tan Pin Siang committed 7 months ago

(security) Fix SSRF in batch runner download_bytes_from_url (#38482)

Juan Pérez de Algaba committed 7 months ago

[Bugfix][Model] Fix PixtralForConditionalGeneration LoRA (#36963)

Jee Jee Li committed 7 months ago

[Model][Quantization] Add GGUF support for MiniMax-M2.1 (#36965)

PikaPikachu committed 7 months ago

[ci] Soft fail and disable retry for AMD build image job (#38505)

Kevin H. Luu committed 7 months ago

[MoE] Add RoutingMethodType.Simulated to TRT-LLM FP8/NVFP4 kernel allowlists (#38329)

Jaewon committed 7 months ago

[Misc] Always use `forward_mulmat` for `Conv3d` on newer versions of torch. (#38487)

Roger Wang committed 7 months ago

[CI] Add temperature=0.0, reduce max_tokens, and add debug prints to audio_in_video tests (#38492)

Andreas Karatzas committed 7 months ago

Add @ZJY0516 to CODEOWNERS (#38497)

Jiangyun Zhu committed 7 months ago

[Bugfix] Support multi-type params parsing for DeepSeek v3.2 (#33703)

Stanislav Kirillov committed 7 months ago

[ROCm][CI] Enable hybrid chunked prefill test (#38317)

Andreas Karatzas committed 7 months ago

[QeRL] Fix online quantized reloading (#38442)

Kyle Sayers committed 7 months ago

[Perf] Remove redundant device copies for CPU-only pooling token IDs, 48.9% E2E throughput improvement (#38139)

Wentao Ye committed 7 months ago

[Transformers v5] fix missing pixtral/voxtral multimodal dispatch (#38410)

allgather committed 7 months ago

[ROCm][CI] Fix cross-attention dispatch for encoder-decoder models (#38450)

Andreas Karatzas committed 7 months ago

[Spec Decode, BugFix] Propagate norm_before_fc from Eagle3 speculator (#38111)

Shubhra Pandit committed 7 months ago

[Feature] Add Qwen3-ForcedAligner support via token classification pooling (#35367)

haosdent committed 7 months ago