Merge pull request #69 from unslothai/fix/auto-detect-lora-in-model-config

fix: auto-detect LoRA adapters for both local and remote HF models in ModelConfig

studio/backend/utils/models/model_config.py

@@ -661,10 +661,44 @@ class ModelConfig:
             identifier = f"unsloth/{identifier}"
             path = identifier
         
+        # Auto-detect LoRA for local paths (check adapter_config.json on disk)
+        if not is_lora and is_local:
+            detected_base = get_base_model_from_lora(path)
+            if detected_base:
+                is_lora = True
+                logger.info(f"Auto-detected local LoRA adapter at '{path}' (base: {detected_base})")
+        
+        # Auto-detect LoRA for remote HF models (check repo file listing)
+        if not is_lora and not is_local:
+            try:
+                from huggingface_hub import model_info as hf_model_info
+                info = hf_model_info(identifier, token=hf_token)
+                repo_files = [s.rfilename for s in info.siblings]
+                if "adapter_config.json" in repo_files:
+                    is_lora = True
+                    logger.info(f"Auto-detected remote LoRA adapter: '{identifier}'")
+            except Exception as e:
+                logger.debug(f"Could not check remote LoRA status for '{identifier}': {e}")
+        
         # Handle LoRA adapters
         base_model = None
         if is_lora:
-            base_model = get_base_model_from_lora(path)
+            if is_local:
+                # Local LoRA: read adapter_config.json from disk
+                base_model = get_base_model_from_lora(path)
+            else:
+                # Remote LoRA: download adapter_config.json from HF
+                try:
+                    from huggingface_hub import hf_hub_download
+                    config_path = hf_hub_download(identifier, "adapter_config.json", token=hf_token)
+                    with open(config_path, 'r') as f:
+                        adapter_config = json.load(f)
+                    base_model = adapter_config.get("base_model_name_or_path")
+                    if base_model:
+                        logger.info(f"Resolved remote LoRA base model: '{base_model}'")
+                except Exception as e:
+                    logger.warning(f"Could not download adapter_config.json for '{identifier}': {e}")
+            
             if not base_model:
                 logger.warning(f"Could not determine base model for LoRA '{path}'")
                 return None

studio/tests/test_remote_lora_detection.py

@@ -0,0 +1,40 @@
+"""
+Test remote LoRA adapter detection via HuggingFace Hub API.
+
+Verifies that we can detect whether a remote HF model is a LoRA adapter
+by checking for adapter_config.json in the repo file listing.
+"""
+import pytest
+from huggingface_hub import model_info
+
+
+def is_remote_lora_adapter(model_id: str, hf_token: str = None) -> bool:
+    """
+    Check if a remote HuggingFace model is a LoRA adapter
+    by looking for adapter_config.json in its repo files.
+    """
+    try:
+        info = model_info(model_id, token=hf_token)
+        filenames = [s.rfilename for s in info.siblings]
+        return "adapter_config.json" in filenames
+    except Exception:
+        return False
+
+
+class TestRemoteLoRADetection:
+    """Test remote LoRA adapter detection via HF Hub API."""
+
+    def test_lora_adapter_detected(self):
+        """edbeeching/llama-se-rl-adapter is a known LoRA adapter on HF."""
+        result = is_remote_lora_adapter("edbeeching/llama-se-rl-adapter")
+        assert result is True, "Expected edbeeching/llama-se-rl-adapter to be detected as a LoRA adapter"
+
+    def test_base_model_not_detected_as_lora(self):
+        """google/gemma-3-4b-it is a full base model, not a LoRA adapter."""
+        result = is_remote_lora_adapter("google/gemma-3-4b-it")
+        assert result is False, "Expected google/gemma-3-4b-it to NOT be detected as a LoRA adapter"
+
+    def test_nonexistent_model_returns_false(self):
+        """A nonexistent model should return False, not raise."""
+        result = is_remote_lora_adapter("this-org-does-not-exist/fake-model-12345")
+        assert result is False, "Expected nonexistent model to return False"