Remove unnecessary `load_weights` methods (#44589)

Harry Mellor committed 4 months ago

[ROCm][CI] Explicitly tear down multimodal offline LLMs (#46999)

Andreas Karatzas committed 4 months ago

Fix docs on main (#47009)

Harry Mellor committed 4 months ago

[MM][CG] Gemma3 Encoder CUDA Graph (#43591)

Soyaazz committed 4 months ago

[Rust Frontend] Add Harmony Renderer for GPT-OSS (#46800)

Bugen Zhao committed 4 months ago

[CPU] Support cpu compressed-tensor w8a8 int8 moe (#42920)

Yuwen Zhou committed 4 months ago

[EPLB] Mask padding in EPLB load recording (#38128)

Ilya Markov committed 4 months ago

Remove boilerplate missed by #46820 (#46956)

Harry Mellor committed 4 months ago

[Spec Decode] Avoid redundant hidden-states gather in draft prefill (#46968)

Woosuk Kwon committed 4 months ago

[Model] Add support for openai/privacy-filter (#41026)

Fabian Joswig committed 4 months ago

[Bugfix][Mooncake] Fix Mooncake lookup prefixes with DCP > 1 (#46855)

Wei Zhao committed 4 months ago

[ROCm][CI] Fix `rlhf_async_new_apis` Example On ROCm (#46895)

Micah Williamson committed 4 months ago

[OCP MX ] Add back emulation to available OCP MX backends list (#46629)

fxmarty-amd committed 4 months ago

[GLM5] Fix minor typo (#46961)

Woosuk Kwon committed 4 months ago

[Bugfix] Transformers backend: recompute `mm_token_type_ids` per request for M-RoPE (#46552)

Gonzague de Carpentier committed 4 months ago

[GLM5] Implement op fusion for GLM5/DSV3.2 (#46876)

Woosuk Kwon committed 4 months ago

[GLM5.2 Perf] Replace MOE all-reduce with reduce-scatter, 3.1%~3.2 E2E Throughput improvement (#46635)

Wentao Ye committed 4 months ago

[ROCm][Perf][MLA] Add AITER FlashAttention MLA prefill backend (`ROCM_AITER_FA`) (#45033)

xaguilar-amd committed 4 months ago

[Bugfix][DSv3.2] Skip indexer weights for index-cache-skipped layers (#46600)

frida-andersson committed 4 months ago

[Render][Speculator] Add return_loss_mask to render endpoint for training data generation (#46846)

Ranran committed 4 months ago