[Model] Introduce verify_and_update_model_config for VerifyAndUpdateConfig. (#31131)

Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
Signed-off-by: wang.yuqi <noooop@126.com>

vllm/config/model.py

@@ -595,7 +595,7 @@ class ModelConfig:
 
         # Avoid running try_verify_and_update_config multiple times
         self.config_updated = False
-
+        self._try_verify_and_update_model_config()
         self._verify_quantization()
         self._verify_cuda_graph()
         self._verify_bnb_config()
@@ -1008,6 +1008,23 @@ class ModelConfig:
                 "when expert parallelism is enabled."
             )
 
+    def _try_verify_and_update_model_config(self):
+        # Avoid running try_verify_and_update_config multiple times
+        if getattr(self, "config_updated", False):
+            return
+
+        architecture = self.architecture
+        if architecture is None:
+            return
+
+        from vllm.model_executor.models.config import (
+            MODELS_CONFIG_MAP,
+        )
+
+        cls = MODELS_CONFIG_MAP.get(architecture, None)
+        if cls is not None:
+            cls.verify_and_update_model_config(self)
+
     def verify_dual_chunk_attention_config(
         self,
         load_config: LoadConfig,

vllm/model_executor/models/config.py

@@ -13,7 +13,7 @@ from vllm.utils.torch_utils import STR_DTYPE_TO_TORCH_DTYPE
 from vllm.v1.kv_cache_interface import FullAttentionSpec, MambaSpec, MLAAttentionSpec
 
 if TYPE_CHECKING:
-    from vllm.config import VllmConfig
+    from vllm.config import ModelConfig, VllmConfig
 
 logger = init_logger(__name__)
 
@@ -21,20 +21,24 @@ logger = init_logger(__name__)
 class VerifyAndUpdateConfig:
     @staticmethod
     def verify_and_update_config(vllm_config: "VllmConfig") -> None:
-        raise NotImplementedError
+        return
+
+    @staticmethod
+    def verify_and_update_model_config(model_config: "ModelConfig") -> None:
+        return
 
 
-class Gemma3TextModelConfig:
+class Gemma3TextModelConfig(VerifyAndUpdateConfig):
     @staticmethod
-    def verify_and_update_config(vllm_config: "VllmConfig") -> None:
-        hf_config = vllm_config.model_config.hf_config
+    def verify_and_update_model_config(model_config: "ModelConfig") -> None:
+        hf_config = model_config.hf_config
         hf_config.is_causal = not hf_config.use_bidirectional_attention
 
 
 class GteNewModelConfig(VerifyAndUpdateConfig):
     @staticmethod
-    def verify_and_update_config(vllm_config: "VllmConfig") -> None:
-        config = vllm_config.model_config.hf_config
+    def verify_and_update_model_config(model_config: "ModelConfig") -> None:
+        config = model_config.hf_config
 
         assert config.__class__.__name__ == "NewConfig"
         assert config.hidden_act == "gelu"
@@ -53,16 +57,15 @@ class GteNewModelConfig(VerifyAndUpdateConfig):
 
 class JambaForSequenceClassificationConfig(VerifyAndUpdateConfig):
     @staticmethod
-    def verify_and_update_config(vllm_config: "VllmConfig") -> None:
-        pooler_config = vllm_config.model_config.pooler_config
+    def verify_and_update_model_config(model_config: "ModelConfig") -> None:
+        pooler_config = model_config.pooler_config
         if pooler_config.use_activation is None:
             pooler_config.use_activation = False
 
 
 class JinaRobertaModelConfig(VerifyAndUpdateConfig):
     @staticmethod
-    def verify_and_update_config(vllm_config: "VllmConfig") -> None:
-        model_config = vllm_config.model_config
+    def verify_and_update_model_config(model_config: "ModelConfig") -> None:
         config = model_config.hf_config
 
         if config.position_embedding_type == "rotary":
@@ -90,10 +93,10 @@ class JinaRobertaModelConfig(VerifyAndUpdateConfig):
 
 class LlamaBidirectionalConfig(VerifyAndUpdateConfig):
     @staticmethod
-    def verify_and_update_config(vllm_config: "VllmConfig") -> None:
+    def verify_and_update_model_config(model_config: "ModelConfig") -> None:
         from vllm.config.pooler import PoolingTypeStr
 
-        hf_config = vllm_config.model_config.hf_config
+        hf_config = model_config.hf_config
         hf_config.is_causal = False
 
         pooling_type_map: dict[str, PoolingTypeStr] = {
@@ -105,7 +108,7 @@ class LlamaBidirectionalConfig(VerifyAndUpdateConfig):
         pooling_type = pooling_type_map.get(hf_config.pooling, None)
         if pooling_type is None:
             raise ValueError(f"pool_type {hf_config.pooling} not supported")
-        vllm_config.model_config.pooler_config.pooling_type = pooling_type
+        model_config.pooler_config.pooling_type = pooling_type
 
 
 class NomicBertModelConfig(VerifyAndUpdateConfig):
@@ -204,8 +207,8 @@ class NomicBertModelConfig(VerifyAndUpdateConfig):
 
 class Qwen2ForProcessRewardModelConfig(VerifyAndUpdateConfig):
     @staticmethod
-    def verify_and_update_config(vllm_config: "VllmConfig") -> None:
-        pooler_config = vllm_config.model_config.pooler_config
+    def verify_and_update_model_config(model_config: "ModelConfig") -> None:
+        pooler_config = model_config.pooler_config
 
         if pooler_config.step_tag_id is None:
             pooler_config.step_tag_id = 151651
@@ -213,8 +216,8 @@ class Qwen2ForProcessRewardModelConfig(VerifyAndUpdateConfig):
 
 class Qwen2ForRewardModelConfig(VerifyAndUpdateConfig):
     @staticmethod
-    def verify_and_update_config(vllm_config: "VllmConfig") -> None:
-        pooler_config = vllm_config.model_config.pooler_config
+    def verify_and_update_model_config(model_config: "ModelConfig") -> None:
+        pooler_config = model_config.pooler_config
 
         if pooler_config.softmax is None:
             pooler_config.softmax = False
@@ -222,8 +225,8 @@ class Qwen2ForRewardModelConfig(VerifyAndUpdateConfig):
 
 class Qwen3ForSequenceClassificationConfig(VerifyAndUpdateConfig):
     @staticmethod
-    def verify_and_update_config(vllm_config: "VllmConfig") -> None:
-        config = vllm_config.model_config.hf_config
+    def verify_and_update_model_config(model_config: "ModelConfig") -> None:
+        config = model_config.hf_config
 
         is_original_qwen3_reranker = getattr(
             config, "is_original_qwen3_reranker", False
@@ -237,23 +240,23 @@ class Qwen3ForSequenceClassificationConfig(VerifyAndUpdateConfig):
             "Try loading the original Qwen3 Reranker?, see: "
             "https://github.com/vllm-project/vllm/tree/main/examples/offline_inference/offline_reranker.py"
         )
-        vllm_config.model_config.hf_config.method = "from_2_way_softmax"
+        model_config.hf_config.method = "from_2_way_softmax"
 
 
 class JinaVLForSequenceClassificationConfig(VerifyAndUpdateConfig):
     @staticmethod
-    def verify_and_update_config(vllm_config: "VllmConfig") -> None:
-        config = vllm_config.model_config.hf_config
+    def verify_and_update_model_config(model_config: "ModelConfig") -> None:
+        config = model_config.hf_config
         config.num_labels = 1
-        pooler_config = vllm_config.model_config.pooler_config
+        pooler_config = model_config.pooler_config
         if pooler_config.logit_bias is None:
             pooler_config.logit_bias = 2.65
 
 
 class SnowflakeGteNewModelConfig(VerifyAndUpdateConfig):
     @staticmethod
-    def verify_and_update_config(vllm_config: "VllmConfig") -> None:
-        config = vllm_config.model_config.hf_config
+    def verify_and_update_model_config(model_config: "ModelConfig") -> None:
+        config = model_config.hf_config
 
         assert config.__class__.__name__ == "GteConfig"
         assert config.hidden_act == "gelu"

vllm/model_executor/models/llama.py

@@ -64,7 +64,6 @@ from .interfaces import (
     SupportsLoRA,
     SupportsPP,
 )
-from .interfaces_base import attn_type
 from .utils import (
     AutoWeightsLoader,
     PPMissingLayer,
@@ -707,14 +706,12 @@ class LlamaForCausalLM(
         return name, loaded_weight
 
 
-@attn_type("encoder_only")
 class LlamaBidirectionalForSequenceClassification(as_seq_cls_model(LlamaForCausalLM)):
     # This class sets the correct attention type and pooling type
     # through LlamaBidirectionalConfig.
     pass
 
 
-@attn_type("encoder_only")
 class LlamaBidirectionalModel(as_embedding_model(LlamaForCausalLM)):
     # This class sets the correct attention type and pooling type
     # through LlamaBidirectionalConfig.