15 Commits
Author SHA1 Message Date
Charles Chen 5ae8d286d2 perf(gemma4): single-launch fused router (topk + softmax + scale) (#26502) 2026-06-02 16:00:17 +08:00
Charles Chen 89ff2bc111 [bug fix] Fix 3 issues when using Gemma4 MTP (#26026) 2026-05-23 03:16:47 -07:00
Charles Chen 847cbada9c Support Gemma4 MoE NVFP4 (#25054) 2026-05-20 22:45:15 -07:00
Charles Chen 78408d2300 Fix frozen kv MTP crash when bonus_tokens is None (#25204) 2026-05-14 15:33:55 -07:00
Charles Chen 12f42f2e7e Support Gemma3/4 + Eagle3 (#23976) 2026-05-09 13:34:56 -07:00
Charles Chen c396e4924b [bug] Fix cache salt and extra keys for prefix cache isolation (#23300) 2026-04-21 13:53:24 -07:00
Charles Chen d22c6a3847 fix: Properly return abort error for streaming requests if the abort is triggered by scheduler (#19357) 2026-03-03 17:18:15 -08:00
af0d35b224 Fix: Reject requests with a duplicate request ID which can cause server crash/hang (#19035)
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: hnyls2002 <lsyincs@gmail.com>
2026-03-03 00:33:25 -08:00
Charles Chen 9e0ef04e5b Support using different attention backend for draft decoding. (#14843) 2025-12-19 08:49:11 +08:00
Charles Chen b051d76dab Fix: add missing get_embed_and_head in MiniMax M2 for Eagle3 (#13297) 2025-11-15 01:36:03 -08:00
Charles Chen f1a9c72de3 Support capturing aux_hidden_states for minimax m2. (#12798) 2025-11-08 01:54:22 -08:00
Charles Chen 659bfd1023 Add GKE's default CUDA runtime lib location to PATH and LD_LIBRARY_PATH. (#8544) 2025-07-30 20:28:07 -07:00
Charles Chen 15f3401343 Fix MTP with Deepseek R1 Fp4 (#7376) 2025-06-23 21:38:07 -07:00
Charles Chen e5ddeb04d5 Quick fix for DeepGemm requant to also cover MTP. (#7378) 2025-06-23 12:08:54 -07:00
Charles Chen 8c16da334e Fix Deepseek R1 0528 FP4 tensor name mismatch issue during weights loading. (#7164) 2025-06-17 11:26:23 -07:00