[ROCm][Build] Add support for AMD Ryzen AI MAX / AI 300 Series (#25908)

Hosang committed about 1 year ago

[BugFix][DP/EP] Fix CUTLASS MLA hang under load (#26026)

Lucas Wilkinson committed about 1 year ago

[CI] Tweaks to GPT-OSS Eval (Blackwell) for stability (#26030)

Michael Goin committed about 1 year ago

Fix test_mamba_ssm_ssd.py due to missing _query_start_loc_to_chunk_indices_offsets (#25995)

Huamin Li committed about 1 year ago

[NVIDIA] Blackwell Family (#24673)

Johnny committed about 1 year ago

[Bugfix] Apply same sampling parameters for both `n=1` and `n>1` (#26005)

Kenichi Maehashi committed about 1 year ago

[Benchmark] Finish documented v0.11.0 deprecation of --endpoint-type (#26007)

Nathan Scott committed about 1 year ago

[MISC] Fix misleading batch_size_capture_list when cuda_graph_sizes < 4 (#25829)

billishyahao committed about 1 year ago

[CI] Only capture a single CUDA graph size in CI by default (#25951)

Harry Mellor committed about 1 year ago

[Misc] Factor out common `_apply_feature_select_strategy` (#26003)

Cyrus Leung committed about 1 year ago

[BugFix][MM] Fix Nonetype error when video is cache in qwen2.5-omni-thinker (#26004)

Wenlong Wang committed about 1 year ago

[Doc] updating torch.compile doc link (#25989)

nadathurv committed about 1 year ago

[Misc]allow disable pynccl (#25421)

Lucia Fang committed about 1 year ago

Update to Transformers `v4.56.2` (#24638)

Harry Mellor committed about 1 year ago

[BugFix] Fix default kv-cache-dtype default for DeepseekV3.2 (#25988)

Lucas Wilkinson committed about 1 year ago

[Bugfix] Fix `__syncwarp` on ROCM (#25996)

Zhewen Li committed about 1 year ago

[MM] Add text-only mode for Qwen3-VL (#26000)

Roger Wang committed about 1 year ago

Fix INT8 quantization error on Blackwell GPUs (SM100+) (#25935)

Param committed about 1 year ago

[Log] Optimize Log for FP8MOE (#25709)

Wentao Ye committed about 1 year ago

Update launch_bounds_utils.h for correct compile on Multiple Cuda Arch - PTXAS out of range Warning (#25843)

Salvatore Cena committed about 1 year ago