Upload folder using huggingface_hub

src/llm.py

@@ -1,12 +1,5 @@
 """
-LLM module. HuggingFace Inference API as primary.
-Works natively from HF Spaces — same infrastructure.
-OpenRouter and Groq as fallback providers.
-
-WHY HF Inference API?
-HF Spaces can always reach HuggingFace's own APIs.
-No network routing issues. Uses existing HF_TOKEN.
-Same Llama 3.3 70B model as others.
+LLM module. OpenRouter as primary, HuggingFace Inference API as fallback.
 """
 
 import os
@@ -17,39 +10,18 @@ from tenacity import retry, stop_after_attempt, wait_exponential
 load_dotenv()
 logger = logging.getLogger(__name__)
 
-# ── HuggingFace Inference API ─────────────────────────────
-_hf_client = None
-
-# ── OpenRouter (free tier, reliable fallback) ──────────────
+# ── OpenRouter (primary) ──────────────────────────────────
 _openrouter_client = None
 
-# ── Groq fallback (works locally, may be blocked on HF Spaces) ──
-_groq_client = None
-
-
-def _init_hf():
-    global _hf_client
-    token = os.getenv("HF_TOKEN")
-    if not token:
-        logger.warning("HF_TOKEN not set — HF Inference API disabled")
-        return False
-    try:
-        from huggingface_hub import InferenceClient
-        _hf_client = InferenceClient(
-            model="meta-llama/Llama-3.3-70B-Instruct",
-            token=token
-        )
-        logger.info("HF Inference API ready (Llama-3.3-70B)")
-        return True
-    except Exception as e:
-        logger.error(f"HF Inference API init failed: {e}")
-        return False
+# ── HuggingFace Inference API (fallback) ──────────────────
+_hf_client = None
 
 
 def _init_openrouter():
     global _openrouter_client
     api_key = os.getenv("OPENROUTER_API_KEY")
     if not api_key:
+        logger.warning("OPENROUTER_API_KEY not set — OpenRouter disabled")
         return False
     try:
         from openai import OpenAI
@@ -57,45 +29,38 @@ def _init_openrouter():
             base_url="https://openrouter.ai/api/v1",
             api_key=api_key,
         )
-        logger.info("OpenRouter ready as fallback")
+        logger.info("OpenRouter ready (primary)")
         return True
     except Exception as e:
         logger.error(f"OpenRouter init failed: {e}")
         return False
 
 
-def _init_groq():
-    global _groq_client
-    api_key = os.getenv("GROQ_API_KEY")
-    if not api_key:
+def _init_hf():
+    global _hf_client
+    token = os.getenv("HF_TOKEN")
+    if not token:
+        logger.warning("HF_TOKEN not set — HF Inference API disabled")
         return False
     try:
-        from groq import Groq
-        _groq_client = Groq(api_key=api_key)
-        logger.info("Groq ready as fallback")
+        from huggingface_hub import InferenceClient
+        _hf_client = InferenceClient(
+            model="meta-llama/Llama-3.3-70B-Instruct",
+            token=token
+        )
+        logger.info("HF Inference API ready (fallback)")
         return True
     except Exception as e:
-        logger.error(f"Groq init failed: {e}")
+        logger.error(f"HF Inference API init failed: {e}")
         return False
 
 
-_hf_ready = _init_hf()
 _openrouter_ready = _init_openrouter()
-_groq_ready = _init_groq()
-
-
-def _call_hf(messages: list) -> str:
-    """Call HuggingFace Inference API."""
-    response = _hf_client.chat_completion(
-        messages=messages,
-        max_tokens=1500,
-        temperature=0.3,
-    )
-    return response.choices[0].message.content
+_hf_ready = _init_hf()
 
 
 def _call_openrouter(messages: list) -> str:
-    """Call OpenRouter free tier."""
+    """Call OpenRouter."""
     response = _openrouter_client.chat.completions.create(
         model="meta-llama/llama-3.3-70b-instruct:free",
         messages=messages,
@@ -105,25 +70,18 @@ def _call_openrouter(messages: list) -> str:
     return response.choices[0].message.content
 
 
-def _call_groq(messages: list) -> str:
-    """Call Groq as fallback."""
-    response = _groq_client.chat.completions.create(
-        model="llama-3.3-70b-versatile",
+def _call_hf(messages: list) -> str:
+    """Call HuggingFace Inference API."""
+    response = _hf_client.chat_completion(
         messages=messages,
+        max_tokens=1500,
         temperature=0.3,
-        max_tokens=1500
     )
     return response.choices[0].message.content
 
 
 def _call_with_fallback(messages: list) -> str:
-    """Try Groq first, then OpenRouter, then HF."""
-    if _groq_ready and _groq_client:
-        try:
-            return _call_groq(messages)
-        except Exception as e:
-            logger.warning(f"Groq failed: {e}, trying OpenRouter")
-
+    """Try OpenRouter first, then HF."""
     if _openrouter_ready and _openrouter_client:
         try:
             return _call_openrouter(messages)