debug: add startup diagnostics to test various models and endpoints

app.py

@@ -578,6 +578,40 @@ async def get_config():
 # Mount static files
 fastapi_app.mount("/static", StaticFiles(directory=str(STATIC_DIR)), name="static")
 
+@fastapi_app.on_event("startup")
+async def run_diagnostics():
+    log.info("Running startup diagnostics for HF Inference API models...")
+    from huggingface_hub import InferenceClient
+    import os
+    token = os.environ.get("HF_TOKEN") or os.environ.get("HUGGINGFACEHUB_API_TOKEN")
+    
+    models = [
+        "Qwen/Qwen2.5-1.5B-Instruct",
+        "HuggingFaceTB/SmolLM2-1.7B-Instruct",
+        "google/gemma-2-2b-it",
+        "microsoft/Phi-3.5-mini-instruct",
+        "microsoft/Phi-4-mini-instruct",
+    ]
+    
+    for m in models:
+        # Test 1: plain text_generation
+        try:
+            client = InferenceClient(model=m, token=token)
+            res = client.text_generation("Say hello", max_new_tokens=10)
+            log.info(f"DIAGNOSTIC: {m} (text_gen) => SUCCESS: {res.strip()}")
+        except Exception as e:
+            log.info(f"DIAGNOSTIC: {m} (text_gen) => FAIL: {str(e)[:150]}")
+            
+        # Test 2: conversational
+        try:
+            client = InferenceClient(model=m, token=token)
+            # Try chat_completion but with explicit hf-inference provider if possible, or just default
+            res = client.chat_completion(messages=[{"role": "user", "content": "Say hello"}], max_tokens=10)
+            log.info(f"DIAGNOSTIC: {m} (chat) => SUCCESS: {res.choices[0].message.content.strip()}")
+        except Exception as e:
+            log.info(f"DIAGNOSTIC: {m} (chat) => FAIL: {str(e)[:150]}")
+
+
 # Mount Gradio app at /gradio — this creates the API + MCP endpoints
 gradio_blocks = create_gradio_app()
 mount_gradio_app(fastapi_app, gradio_blocks, path="/gradio")