[Perf][MRV2] Reuse Mamba/GDN metadata across KV cache groups (#58762)

Luca Motz committed 24 days ago

[Minimax-M3] Add Encoder CUDA graph support (#58673)

Thien Tran committed 24 days ago

[Bugfix][Frontend] Apply Harmony adjust_request in batched chat completions (#58958)

Flora Feng committed 24 days ago

[Model] Enable LoRA support for RobertaForSequenceClassification (#58884)

jz-yolo committed 24 days ago

[Test][ROCm] Stabilize the mixed OLMoE LoRA test (#58945)

Andreas Karatzas committed 24 days ago

[CPU] Build CPU wheels on Ubuntu 22.04 with AMX-FP8 support (#58515)

zhao, zhenhui committed 24 days ago

[Bugfix][EPD] Skip sampling for encoder-only async steps (#58490)

Tianyu Guo committed 24 days ago

[Bugfix][Frontend] Use a fresh parser per choice in non-streaming chat completions (#58939)

Flora Feng committed 24 days ago

[Bugfix][Frontend] Sample batched chat completions from the adjusted requests (#58929)

Flora Feng committed 24 days ago

[Mypy] Fix mypy typing for Ultravox and Unlimited-OCR models (#58239)

Ashraf Bhuiyan committed 24 days ago

[KimiViT][Perf] Fuse per-layer QK RoPE into one in-place kernel (#58651)

Thien Tran committed 24 days ago

[Bugfix][Frontend] Count Responses reasoning tokens per tool round (#58927)

Flora Feng committed 24 days ago

[ROCm] Bump AITER to v0.1.23 (#58867)

Fangzhou Ai committed 24 days ago

[Frontend] Switch Python Harmony dependency to oss-harmony (#55128)

Anton Peganov committed 24 days ago

[Bugfix][KV Connector] Retry Mooncake bootstrap registration on timeout (reopens #55763) (#58919)

Chao Lei committed 24 days ago

[ROCm][Bugfix] Fall back to default GEMM for CPU tensors on ROCm builds (#58923)

Fangzhou Ai committed 24 days ago

[Perf][MoE] Use fused MiniMax2 routing with non-unit routed scaling (#58880)

Jiangyun Zhu committed 24 days ago

[Fast Start] Add `/health` endpoint for the weight cache daemon (#58552)

Xun Sun committed 24 days ago

[Bugfix] Support repsonse_format + tool_choice=auto (#56086)

Artem Perevedentsev committed 24 days ago

[Perf][Pooling] Avoid blocking seq_lens GPU-to-CPU copy for pooling in FlashInfer metadata builder (#57214)

Frank Wang committed 24 days ago