[Bugfix] Don't attempt to use triton if no driver is active (#19561)

Konrad Zawora committed over 1 year ago

Only build CUTLASS MoE kernels on Hopper (#19648)

Huy Do committed over 1 year ago

[Hardware][NVIDIA][kernel] Fp4 MOE quant kernel optimization (#19500)

jiahanc committed over 1 year ago

[Bugfix] Fix auto dtype casting for BatchFeature (#19316)

Isotr0py committed over 1 year ago

[Misc] Modularize CLI Argument Parsing in Benchmark Scripts (#19593)

Reid committed over 1 year ago

[Bugfix] Fix the speculative decoding test by setting the target dtype (#19633)

Lu Fang committed over 1 year ago

[V1][Metrics] Deprecate metrics with gpu_ prefix for non GPU specific metrics. (#18354)

Saheli Bhattacharjee committed over 1 year ago

[BugFix] Fix DP Coordinator incorrect debug log message (#19624)

Nick Hill committed over 1 year ago

Adding "AMD: Multi-step Tests" to amdproduction. (#19508)

Concurrensee committed over 1 year ago

[torch.compile] Use custom ops when use_inductor=False (#19618)

Woosuk Kwon committed over 1 year ago

[Doc] Add troubleshooting section to k8s deployment (#19377)

Anna Pendleton committed over 1 year ago

[CUDA] Enable full cudagraph for FlashMLA (#18581)

Luka Govedič committed over 1 year ago

[doc][mkdocs] fix the duplicate Supported features sections in GPU docs (#19606)

Reid committed over 1 year ago

[Refactor] Remove unused variables in `moe_permute_unpermute_kernel.inl` (#19573)

Wentao Ye committed over 1 year ago

[Model] Fix minimax model cache & lm_head precision (#19592)

qscqesze committed over 1 year ago

[BugFix] Honor `enable_caching` in connector-delayed kvcache load case (#19435)

Nick Hill committed over 1 year ago

[torch.compile] reorganize the cache directory to support compiling multiple models (#19064)

youkaichao committed over 1 year ago

use base version for version comparison (#19587)

Boyuan Feng committed over 1 year ago

[Fix] The zip function in Python 3.9 does not have the strict argument (#19549)

汪志鹏 committed over 1 year ago

[Fix] bump mistral common to support magistral (#19533)

汪志鹏 committed over 1 year ago