[2/2] Support Qserve (#6521)

python/sglang/srt/configs/model_config.py

@@ -349,6 +349,7 @@ class ModelConfig:
             "w8a8_int8",
             "w8a8_fp8",
             "moe_wna16",
+            "qoq",
         ]
         compatible_quantization_methods = {
             "modelopt_fp4": ["modelopt"],
@@ -458,6 +459,8 @@ def _get_and_verify_dtype(
     # NOTE: getattr(config, "torch_dtype", torch.float32) is not correct
     # because config.torch_dtype can be None.
     config_dtype = getattr(config, "torch_dtype", None)
+    if isinstance(config_dtype, str):
+        config_dtype = _STR_DTYPE_TO_TORCH_DTYPE.get(config_dtype, None)
     if config_dtype is None:
         config_dtype = torch.float32