[FIX] trtllm-moe-fp4-renorm for Qwen series models (#14350)
This commit is contained in:
@@ -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")
|
||||
|
||||
Reference in New Issue
Block a user