minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

Mistral Large 3 NVFP4 TRTLLM MoE support (#15049)

Browse Source
This commit is contained in:
elvischenv
2025-12-18 11:11:42 +08:00
committed by GitHub
parent 9e7656be80
commit 9970ee34e8
7 changed files with 340 additions and 151 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/server_args.py
+2 -1
View File
@@ -1790,8 +1790,9 @@ class ServerArgs:
"modelopt_fp4",
"fp8",
"modelopt_fp8",
"compressed-tensors",
None,
], f"Invalid quantization '{self.quantization}'. \nFlashInfer TRTLLM MOE supports only: 'modelopt_fp4', 'fp8', 'modelopt_fp8', or bfloat16 (None)."
], f"Invalid quantization '{self.quantization}'. \nFlashInfer TRTLLM MOE supports only: 'modelopt_fp4', 'fp8', 'modelopt_fp8', 'compressed-tensors', or bfloat16 (None)."
self.disable_shared_experts_fusion = True
logger.warning(
"FlashInfer TRTLLM MoE is enabled. --disable-shared-experts-fusion is automatically set."