🐛 Bugfix: Fixed an issue with batch model addition and validation in Alibaba Bailian. (#3350)

Author: YehongPan

backend/services/providers/dashscope_provider.py

@@ -71,15 +71,19 @@ def _is_dashscope_video_understanding_model(model_id: str, desc: str, req_mods:
 
 
 def _is_dashscope_image_understanding_model(model_id: str, desc: str, req_mods: set, res_mods: set) -> bool:
-    searchable_text = f"{model_id} {desc.lower()}"
-    if _is_dashscope_image_generation_model(model_id, desc, req_mods, res_mods):
-        return False
-    if _is_dashscope_video_understanding_model(model_id, desc, req_mods, res_mods):
-        return False
-    if ("image" in req_mods or "video" in req_mods) and "text" in res_mods:
+    """Determine if model is for image understanding.
+
+    A model qualifies for image understanding if:
+    - Its output is text AND input contains image or video (multimodal image understanding), OR
+    - Model ID explicitly matches image-understanding keywords.
+
+    Note: This check is independent of video/image-generation classification.
+    A model that is also a video model (vlm3) can simultaneously be an image model (vlm).
+    """
+    if "text" in res_mods and ("image" in req_mods or "video" in req_mods):
         return True
     return _is_dashscope_explicit_image_understanding_model(model_id) or _has_keyword(
-        searchable_text, DASHSCOPE_IMAGE_UNDERSTANDING_KEYWORDS
+        model_id, DASHSCOPE_IMAGE_UNDERSTANDING_KEYWORDS
     )
 
 
@@ -142,9 +146,10 @@ async def get_models(self, provider_config: Dict) -> List[Dict]:
                 "stt": []  # Maps to "stt"
             }
 
-            # Classify models and inject canonical fields expected downstream
+            # Classify models into multiple buckets based on capabilities.
+            # A model can belong to more than one bucket (e.g., a model that
+            # accepts Image+Video+Text and returns Text is both vlm3, vlm, and llm).
             for model_obj in all_models:
-                # Extract key fields for logical determination (lowercased for robustness)
                 m_id = model_obj.get('model', '').lower()
                 desc = model_obj.get('description', '')
                 metadata = model_obj.get('inference_metadata') or {}
@@ -188,26 +193,28 @@ async def get_models(self, provider_config: Dict) -> List[Dict]:
                     categorized_models['tts'].append(cleaned_model)
                     continue
 
-                # 5. VLM
-                if _is_dashscope_video_understanding_model(m_id, desc, req_mods, res_mods):
-                    cleaned_model.update({"model_tag": "chat", "model_type": "vlm3"})
-                    categorized_models['vlm3'].append(cleaned_model)
-                    continue
-
+                # Multimodal / general-purpose types - non-exclusive, no continue.
+                # Each check is independent; one model can qualify for multiple buckets.
                 if _is_dashscope_image_generation_model(m_id, desc, req_mods, res_mods):
-                    cleaned_model.update({"model_tag": "chat", "model_type": "vlm2"})
-                    categorized_models['vlm2'].append(cleaned_model)
-                    continue
+                    vlm2_model = cleaned_model.copy()
+                    vlm2_model.update({"model_tag": "chat", "model_type": "vlm2"})
+                    categorized_models['vlm2'].append(vlm2_model)
+
+                if _is_dashscope_video_understanding_model(m_id, desc, req_mods, res_mods):
+                    vlm3_model = cleaned_model.copy()
+                    vlm3_model.update({"model_tag": "chat", "model_type": "vlm3"})
+                    categorized_models['vlm3'].append(vlm3_model)
 
                 if _is_dashscope_image_understanding_model(m_id, desc, req_mods, res_mods):
-                    cleaned_model.update({"model_tag": "chat", "model_type": "vlm"})
-                    categorized_models['vlm'].append(cleaned_model)
-                    continue
+                    vlm_model = cleaned_model.copy()
+                    vlm_model.update({"model_tag": "chat", "model_type": "vlm"})
+                    categorized_models['vlm'].append(vlm_model)
 
-                # 6. Chat / LLM
+                # Fallback to llm when no specialized modality is present
                 if 'Text' in req_mod or 'Text' in res_mod:
-                    cleaned_model.update({"model_tag": "chat", "model_type": "llm"})
-                    categorized_models['chat'].append(cleaned_model)
+                    llm_model = cleaned_model.copy()
+                    llm_model.update({"model_tag": "chat", "model_type": "llm"})
+                    categorized_models['chat'].append(llm_model)
 
             # Return the specific list based on the requested target_model_type
             if target_model_type == "llm":

frontend/hooks/model/useDashscopeModelList.ts

@@ -39,9 +39,7 @@ export const useDashscopeModelList = ({
     const modelType =
       form.type === "embedding" && form.isMultimodal
         ? ("multi_embedding" as ModelType)
-        : form.type === "vlm2" || form.type === "vlm3"
-          ? ("vlm" as ModelType)
-          : form.type;
+        : form.type;
 
     try {
       // Use manage interface if tenantId is provided (for super admin)