[FIX] trtllm-moe-fp4-renorm for Qwen series models (#14350)

This commit is contained in:
Sam
2025-12-05 04:52:21 +08:00
committed by GitHub
parent d8faf2f361
commit 922756aaa1
2 changed files with 5 additions and 2 deletions

View File

@@ -1261,7 +1261,6 @@ class ServerArgs:
)
self.disable_overlap_schedule = True
if is_sm100_supported():
self.attention_backend = "triton"
quantization_config = getattr(hf_config, "quantization_config", None)
quant_method = (
quantization_config.get("quant_method")