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

fix server args bug (#14725)

Browse Source
This commit is contained in:
TomerBN-Nvidia
2025-12-09 21:05:05 -08:00
committed by GitHub
parent b0f531ad28
commit b1cbfce612
1 changed files with 6 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/server_args.py
+6 -5
View File
@@ -1233,21 +1233,22 @@ class ServerArgs:
) )
self.disable_radix_cache = True self.disable_radix_cache = True
elif model_arch in ["NemotronHForCausalLM"]: elif model_arch in ["NemotronHForCausalLM"]:
if self.model_config.quantization in [ model_config = self.get_model_config()
if model_config.quantization in [
"modelopt", "modelopt",
"modelopt_fp8", "modelopt_fp8",
"modelopt_fp4", "modelopt_fp4",
]: ]:
assert self.model_config.hf_config.mlp_hidden_act == "relu2" assert model_config.hf_config.mlp_hidden_act == "relu2"
if self.model_config.quantization == "modelopt": if model_config.quantization == "modelopt":
self.quantization = ( self.quantization = (
"modelopt_fp4" "modelopt_fp4"
if self.model_config.hf_config.quantization_config["quant_algo"] if model_config.hf_config.quantization_config["quant_algo"]
== "NVFP4" == "NVFP4"
else "modelopt_fp8" else "modelopt_fp8"
) )
else: else:
self.quantization = self.model_config.quantization self.quantization = model_config.quantization
self.moe_runner_backend = "flashinfer_cutlass" self.moe_runner_backend = "flashinfer_cutlass"
elif model_arch in [ elif model_arch in [
"Qwen3MoeForCausalLM", "Qwen3MoeForCausalLM",