Add ModelOpt Qwen3 nvfp4 support (#20101)

Signed-off-by: Zhiyu Cheng <zhiyuc@nvidia.com>

vllm/model_executor/model_loader/weight_utils.py

@@ -764,39 +764,41 @@ def maybe_remap_kv_scale_name(name: str, params_dict: dict) -> Optional[str]:
             return None
         return remapped_name
 
-    possible_scale_names = [".k_scale", ".v_scale"]
+    # Define scale name mapping patterns in order of precedence
-    modelopt_scale_names = [
+    scale_mapping_patterns = [
-        ".self_attn.k_proj.k_scale", ".self_attn.v_proj.v_scale"
+        # ModelOpt format: .self_attn.{k,v}_proj.{k,v}_scale ->
+        # .self_attn.attn.{k,v}_scale
+        (r"\.self_attn\.([kv])_proj\.([kv])_scale$",
+         r".self_attn.attn.\2_scale"),
+        # QKV proj format: .self_attn.qkv_proj.{k,v}_scale ->
+        # .self_attn.attn.{k,v}_scale
+        (r"\.self_attn\.qkv_proj\.([kv])_scale$", r".self_attn.attn.\1_scale"),
+        # Qwen3 MoE format: .self_attn.qkqkv_proj.{k,v}_scale ->
+        # .self_attn.attn.{k,v}_scale
+        (r"\.self_attn\.qkqkv_proj\.([kv])_scale$", r".self_attn.attn.\1_scale"
+         ),
+        # Default format: .{k,v}_scale -> .attn.{k,v}_scale
+        (r"\.([kv])_scale$", r".attn.\1_scale"),
     ]
-    # Also support qkv_proj scale parameters (from stacked parameter processing)
+
-    qkv_proj_scale_names = [
+    # Check if name ends with k_scale or v_scale
-        ".self_attn.qkv_proj.k_scale", ".self_attn.qkv_proj.v_scale"
+    if name.endswith((".k_scale", ".v_scale")):
-    ]
+        import regex as re
-    for scale_name in possible_scale_names:
+
-        if name.endswith(scale_name):
+        for pattern, replacement in scale_mapping_patterns:
-            if any(mo_scale_name in name
+            if re.search(pattern, name):
-                   for mo_scale_name in modelopt_scale_names):
+                remapped_name = re.sub(pattern, replacement, name)
-                remapped_name = name.replace(
+                if remapped_name not in params_dict:
-                    f".self_attn.{scale_name[1]}_proj{scale_name}",
+                    scale_type = name.split(".")[-1]
-                    f".self_attn.attn{scale_name}")
+                    logger.warning_once(
-            elif any(qkv_scale_name in name
+                        "Found %s in the checkpoint (e.g. %s), but not found the expected name in the model (e.g. %s). %s is not loaded.",  # noqa: E501
-                     for qkv_scale_name in qkv_proj_scale_names):
+                        scale_type,
-                # Handle qkv_proj scale parameters
+                        name,
-                remapped_name = name.replace(
+                        remapped_name,
-                    f".self_attn.qkv_proj{scale_name}",
+                        scale_type,
-                    f".self_attn.attn{scale_name}")
+                    )
-            else:
+                    return None
-                remapped_name = name.replace(scale_name, f".attn{scale_name}")
+                return remapped_name
-            if remapped_name not in params_dict:
-                logger.warning_once(
-                    "Found %s in the checkpoint (e.g. %s), but not found the expected name in the model (e.g. %s). %s is not loaded.",  # noqa: E501
-                    scale_name,
-                    name,
-                    remapped_name,
-                    scale_name,
-                )
-                return None
-            return remapped_name
 
     # If there were no matches, return the untouched param name
     return name

vllm/model_executor/models/qwen2.py

@@ -408,9 +408,18 @@ class Qwen2Model(nn.Module):
                     continue
                 if is_pp_missing_parameter(name, self):
                     continue
+                if name.endswith("scale"):
+                    # Remapping the name of FP8 kv-scale.
+                    name = maybe_remap_kv_scale_name(name, params_dict)
+                    if name is None:
+                        continue
                 param = params_dict[name]
-                weight_loader = param.weight_loader
+                weight_loader = getattr(param, "weight_loader",
-                weight_loader(param, loaded_weight, shard_id)
+                                        default_weight_loader)
+                if weight_loader == default_weight_loader:
+                    weight_loader(param, loaded_weight)
+                else:
+                    weight_loader(param, loaded_weight, shard_id)
                 break
             else:
                 # Skip loading extra bias for GPTQ models.

vllm/model_executor/models/qwen3_moe.py

@@ -48,7 +48,8 @@ from vllm.model_executor.layers.quantization import QuantizationConfig
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead, VocabParallelEmbedding)
-from vllm.model_executor.model_loader.weight_utils import default_weight_loader
+from vllm.model_executor.model_loader.weight_utils import (
+    default_weight_loader, maybe_remap_kv_scale_name)
 from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 
@@ -471,12 +472,21 @@ class Qwen3MoeModel(nn.Module):
                 # Skip layers on other devices.
                 if is_pp_missing_parameter(name, self):
                     continue
+                if name.endswith("scale"):
+                    # Remapping the name of FP8 kv-scale.
+                    name = maybe_remap_kv_scale_name(name, params_dict)
+                    if name is None:
+                        continue
                 if name not in params_dict:
                     continue
 
                 param = params_dict[name]
-                weight_loader = param.weight_loader
+                weight_loader = getattr(param, "weight_loader",
-                weight_loader(param, loaded_weight, shard_id)
+                                        default_weight_loader)
+                if weight_loader == default_weight_loader:
+                    weight_loader(param, loaded_weight)
+                else:
+                    weight_loader(param, loaded_weight, shard_id)
                 break
             else:
                 is_expert_weight = False