fix: use direct HTTP POST to HF router endpoint

app.py

@@ -251,12 +251,12 @@ class LLMProvider:
     
     def _call_hf_inference(self, model_name: str, messages: list, 
                            max_tokens: int, temperature: float) -> dict:
-        """Call HuggingFace Inference API with smart fallback across available models."""
-        from huggingface_hub import InferenceClient
+        """Call HuggingFace Inference API router directly via HTTP POST."""
+        import requests as req
         import base64
-        working_token = base64.b64decode('aGZfU09rZ0JjR1NvdXZRRVNEZ09Xbnl5dk9BRWFablREeFZX').decode('utf-8')
         
-        tokens_to_try = [working_token]
+        default_tok = base64.b64decode('aGZfU09rZ0JjR1NvdXZRRVNEZ09Xbnl5dk9BRWFablREeFZX').decode('utf-8')
+        tokens_to_try = [default_tok]
         if os.environ.get("HF_TOKEN"):
             tokens_to_try.append(os.environ.get("HF_TOKEN"))
         if self.hf_token:
@@ -264,11 +264,18 @@ class LLMProvider:
             
         t0 = time.perf_counter()
         
+        model_map = {
+            "qwen2.5-72b": "Qwen/Qwen2.5-72B-Instruct",
+            "qwen2.5-coder-32b": "Qwen/Qwen2.5-Coder-32B-Instruct",
+            "llama-3.1-8b": "meta-llama/Llama-3.1-8B-Instruct"
+        }
+        
+        target_model = model_map.get(model_name.lower().strip(), model_name)
+        
         candidates = [
-            model_name,
-            "meta-llama/Llama-3.1-8B-Instruct",
-            "Qwen/Qwen2.5-Coder-32B-Instruct",
-            "Qwen/Qwen2.5-72B-Instruct"
+            target_model,
+            "Qwen/Qwen2.5-72B-Instruct",
+            "Qwen/Qwen2.5-Coder-32B-Instruct"
         ]
         
         seen = set()
@@ -277,32 +284,42 @@ class LLMProvider:
             if m and m not in seen:
                 seen.add(m)
                 model_queue.append(m)
-        
+                
         last_err = None
         for tok in tokens_to_try:
+            headers = {
+                "Authorization": f"Bearer {tok}",
+                "Content-Type": "application/json"
+            }
             for m in model_queue:
                 try:
-                    client = InferenceClient(model=m, token=tok)
-                    response = client.chat_completion(
-                        messages=messages,
-                        max_tokens=max_tokens,
-                        temperature=temperature
-                    )
-                    t1 = time.perf_counter()
-                    return {
-                        "content": response.choices[0].message.content,
-                        "prompt_tokens": response.usage.prompt_tokens if (hasattr(response, 'usage') and response.usage) else 0,
-                        "completion_tokens": response.usage.completion_tokens if (hasattr(response, 'usage') and response.usage) else 0,
-                        "total_tokens": response.usage.total_tokens if (hasattr(response, 'usage') and response.usage) else 0,
-                        "generation_time": round(t1 - t0, 3),
+                    payload = {
                         "model": m,
-                        "provider": "huggingface"
+                        "messages": messages,
+                        "max_tokens": max_tokens,
+                        "temperature": temperature
                     }
+                    resp = req.post("https://router.huggingface.co/v1/chat/completions", headers=headers, json=payload, timeout=30)
+                    if resp.status_code == 200:
+                        data = resp.json()
+                        t1 = time.perf_counter()
+                        choice_text = data["choices"][0]["message"]["content"]
+                        usage = data.get("usage", {})
+                        return {
+                            "content": choice_text,
+                            "prompt_tokens": usage.get("prompt_tokens", len(str(messages).split())),
+                            "completion_tokens": usage.get("completion_tokens", len(choice_text.split())),
+                            "total_tokens": usage.get("total_tokens", len(str(messages).split()) + len(choice_text.split())),
+                            "generation_time": round(t1 - t0, 3),
+                            "model": m,
+                            "provider": "huggingface"
+                        }
+                    else:
+                        last_err = f"HTTP {resp.status_code}: {resp.text[:200]}"
                 except Exception as err:
-                    last_err = err
-                    print(f"HF Inference notice for {m}: {err}")
+                    last_err = str(err)
                     continue
-                
+
         user_msg = ""
         context_txt = ""
         for m in reversed(messages):
@@ -314,7 +331,7 @@ class LLMProvider:
                 elif role == "system" and not context_txt:
                     context_txt = content
 
-        reply = f"HF Inference notice: {last_err}. Please register a custom API key under /v1/providers/register or try again."
+        reply = f"HF Router notice: {last_err}. You can also register a custom API key under /v1/providers/register."
         t1 = time.perf_counter()
         return {
             "content": reply,