Feature/Add GET endpoint to query loaded LoRA adapters (#12229)

python/sglang/srt/entrypoints/http_server.py

@@ -1168,6 +1168,8 @@ async def available_models():
     """Show available models. OpenAI-compatible endpoint."""
     served_model_names = [_global_state.tokenizer_manager.served_model_name]
     model_cards = []
+
+    # Add base model
     for served_model_name in served_model_names:
         model_cards.append(
             ModelCard(
@@ -1176,6 +1178,20 @@ async def available_models():
                 max_model_len=_global_state.tokenizer_manager.model_config.context_len,
             )
         )
+
+    # Add loaded LoRA adapters
+    if _global_state.tokenizer_manager.server_args.enable_lora:
+        lora_registry = _global_state.tokenizer_manager.lora_registry
+        for _, lora_ref in lora_registry.get_all_adapters().items():
+            model_cards.append(
+                ModelCard(
+                    id=lora_ref.lora_name,
+                    root=lora_ref.lora_path,
+                    parent=served_model_names[0],
+                    max_model_len=None,
+                )
+            )
+
     return ModelList(data=model_cards)
 
 

python/sglang/srt/entrypoints/openai/protocol.py

@@ -54,6 +54,7 @@ class ModelCard(BaseModel):
     created: int = Field(default_factory=lambda: int(time.time()))
     owned_by: str = "sglang"
     root: Optional[str] = None
+    parent: Optional[str] = None
     max_model_len: Optional[int] = None
 
 

python/sglang/srt/lora/lora_registry.py

@@ -205,3 +205,12 @@ class LoRARegistry:
         Returns the total number of LoRA adapters currently registered.
         """
         return len(self._registry)
+
+    def get_all_adapters(self) -> Dict[str, LoRARef]:
+        """
+        Returns a dictionary of all registered LoRA adapters.
+
+        Returns:
+            Dict[str, LoRARef]: A dictionary mapping LoRA names to LoRARef objects.
+        """
+        return dict(self._registry)