[ModelRunner V2] Fix spec decoding + logprobs (#33391)

Nick Hill committed 9 months ago

[Attention] Move MLA `forward` from backend to layer (#33284)

Matthew Bonanni committed 9 months ago

[Deprecation] Deprecate `seed_everything` and `scatter_mm_placeholders` in v0.15 (#33362)

Wentao Ye committed 9 months ago

[Bugfix] Fix typo in read_offset variable name (#33426)

Alberto Ferrer committed 9 months ago

[Bugfix][ROCm] Fixing the skinny gemm dispatch logic from #32831 (#33366)

Gregory Shtrasberg committed 9 months ago

Refactor NVFP4 Linear utils for ModelOpt and CT (#33201)

Michael Goin committed 9 months ago

[CI][HPU]accelerate hpu test by skip python re-install and clean container name (#33286)

Chendi.Xue committed 9 months ago

Indicate compile mode in the benchmark results (#32990)

Huy Do committed 9 months ago

[Hardware][SM100] Add TRTLLM Kernel for INT4 W4A16 Kernel. (#32437)

Pavani Majety committed 9 months ago

[Quantization][ROCm] Fix MoE weight loading to be robust (Qwen3_MoE/Qwen3_next as example models) (#33173)

xuebwang-amd committed 9 months ago

fix QERL attention import path (#33432)

Vasiliy Kuznetsov committed 9 months ago

[Kernel] [Helion] [1/N] Add Helion ConfigManager (#32740)

Yanan Cao committed 9 months ago

Fix encoder-decoder model disabling mm processor cache (#33236)

Harry Mellor committed 9 months ago

[CI] Qwen3-ASR transcriptios tests (#33414)

Nicolò Lucchesi committed 9 months ago

Support FP8 block quant for CompressedTensorsW8A16Fp8 (#33280)

Michael Goin committed 9 months ago

[QeRL] Layerwise Reloading (#32133)

Kyle Sayers committed 9 months ago

[BugFix][LoRA] TritonExperts is ModularMoEPath for FP8 models (#33393)

Danielle Robinson committed 9 months ago

Disable Cascade Attention for Batch Invariance (#32561)

Frank Wang committed 9 months ago

Improve Mistral format checks. (#33253)

Julien Denize committed 9 months ago

Fix `test_moe.py` for Transformers v5 (#33413)

Harry Mellor committed 9 months ago