[ROCm] mori: add InterNodeV1LL inter-node kernel selection via VLLM_MORI_INTERNODE_KERNEL (#41751)

jatseng-ai committed 5 months ago

Fix early CUDA init (#43791)

Harry Mellor committed 5 months ago

[Docs] Fix the duplicate doc icon issue (#43546)

Chunyang Wen committed 5 months ago

[BugFix] HFValidationError with cloud storage URIs when HF_HUB_OFFLINE=1 (#39155)

Injae Ryou committed 5 months ago

[misc] Bump cutedsl version to 4.5.2 (#43745)

Yongye Zhu committed 5 months ago

[Bugfix] Map reasoning_effort to enable_thinking in chat template kwargs (#43401)

Ashwin Giridharan committed 5 months ago

[Kernel] Enable TritonW4A16LinearKernel as CUDA fallback for non-Marlin-aligned W4A16 shapes (#43731)

Luciano Martins committed 5 months ago

[Docs] Fix MLA prefill backend default docs (#43697)

Mohammad Miadh Angkad committed 5 months ago

[Rust Frontend] Align tool parser fallback behavior between streaming & non-streaming paths (#43662)

Bugen Zhao committed 5 months ago

[Doc] Add Ascend NPU tab to the quickstart installation guide (#43550)

Aditya Singh committed 5 months ago

[ROCm][GPT-OSS] Avoid repeated compile-time `cos_sin_cache.to(bf16)` casts in rotary path (#42833)

akii96 committed 5 months ago

[Frontend] Add MiniCPM5 XML tool call parser (#43175)

zhangtao2-1 committed 5 months ago

[MRV2][BugFix] Fix KV connector handling in spec decode case (#43719)

Nick Hill committed 5 months ago

[ROCm][Perf] Expose AITER MoE sorting dispatch policy via env var (#39177)

Nico Holmberg committed 5 months ago

[KVConnector][Mooncake] Wire reset_cache cascade end-to-end (#42694)

aoshen02 committed 5 months ago

Fix test_aot_compile for torch 2.12 (#43695)

Angela Yi committed 5 months ago

[DSv4] Refactor compressor & Fix ROCm compatibility (#43710)

Woosuk Kwon committed 5 months ago

[Deprecation] Deprecate functions as scheduled for v0.21.0 (#43358)

Wentao Ye committed 5 months ago

[MLA][Attention] Add OOT MLA prefill backend registration mechanism (#43325)

Matthew Bonanni committed 5 months ago

[Attention] Make FlexAttention and FlashAttention use num-blocks first layouts (#42095)

Lucas Wilkinson committed 5 months ago