A high-throughput and memory-efficient inference and serving engine for LLMs
main
[BUGFIX] fix ovis2_5 multimodal tokens (#52623)
Miłosz Grunwald committed 27 days ago
[PD][PushConnector] Record last activity of remotes on the D side (#52245)
snadampal committed 27 days ago
[CI] Report to CRCR after all jobs finish, gated on the build's long pole (#58628)
Andrey Talman committed 27 days ago
[XPU][CI] enable prompt embeds tests on XPU (#58283)
Fanli Lin committed 27 days ago
[Bugfix][KV Cache] Fix incremental multimodal block hashing (#51694)
Jellow committed 27 days ago
[Bugfix][Outlines] Fix EOS termination and unconstrained masks after rejected drafts (#58612)
Nick Hill committed 27 days ago
[DFlash] Capture the context K/V precompute in the draft CUDA graph (#57632)
Juntian Liu committed 27 days ago
[ROCm][CI] skip the ROCm MRV1 default where MRV1 cannot serve the config (#58535)
stefankoncarevic committed 27 days ago
[Bugfix] Pass quant_config to DiffusionGemma's ParallelLMHead (#48521)
Aaron Kang committed 27 days ago
[Perf] DiffusionGemma: constrained reads over the request's logprob_token_ids (#58216)
Matt Mastracci committed 27 days ago
[ROCm] Fix misrouting race-condition in multi-decode P/D disagg with mori-io (#51681)
Vincent Cavé committed 28 days ago
[Cleanup] Remove online quantization support in `fp8.py` in favor of online shorthands (#53585)
fxmarty-amd committed 28 days ago
[Refactor] Move auxiliary files out of the repository root (#58572)
Misha Goin committed 28 days ago
[Bugfix][Quantization] Add Humming to the W4A8 (INT4xFP8) MoE oracle (#58427)
Misha Goin committed 28 days ago
[Fast Start] Wait for weight cache daemon readiness (#58370)
Xun Sun committed 28 days ago
[Core] Skip JIT monitor when JIT warmup is disabled (#58590)
Nick Hill committed 28 days ago
[Bugfix] Keep JIT warmup under enforce-eager when fault tolerance is on (#58593)
Nick Hill committed 28 days ago
[Bugfix] Count unsplit Idefics3 image patches (#48760)
nightcityblade committed 28 days ago
[transformer] RMSNorm matching for alternative rsqrt (#54461)
Thomas Ortner committed 28 days ago
Revert "[DSpark] Support pipeline-parallel targets in aggregated serving (#56956)" (#58484)
Nick Hill committed 28 days ago