added hugging face inference

config.py

@@ -8,9 +8,9 @@ from crewai import LLM
 NBA_DATA_PATH = "nba24-25.csv"
 
 # LLM Configuration - Choose your provider
-# Options: "openai", "ollama", "litellm", "openrouter"
-# Default to openrouter for free tier access
-LLM_PROVIDER = os.getenv("LLM_PROVIDER", "openrouter")  # Default to free tier
+# Options: "openai", "ollama", "litellm", "openrouter", "huggingface"
+# Default to huggingface for best open-source model
+LLM_PROVIDER = os.getenv("LLM_PROVIDER", "huggingface")
 
 # OpenAI Configuration
 OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
@@ -24,11 +24,13 @@ OLLAMA_MODEL = os.getenv("OLLAMA_MODEL", "llama3.2")  # Options: llama3.2, mistr
 LITELLM_MODEL = os.getenv("LITELLM_MODEL", "huggingface/meta-llama/Llama-3.2-3B-Instruct")
 LITELLM_API_KEY = os.getenv("LITELLM_API_KEY", "")  # Optional, depends on provider
 
-# OpenRouter Configuration (access to many open-source models)
+# Hugging Face Configuration (for using HF Inference API)
+HF_API_KEY = os.getenv("HF_API_KEY", "")  # Get from https://huggingface.co/settings/tokens
+HF_MODEL = os.getenv("HF_MODEL", "meta-llama/Llama-3.1-8B-Instruct")  # Best open-source model (8B params)
+
+# OpenRouter Configuration (backup option)
 OPENROUTER_API_KEY = os.getenv("OPENROUTER_API_KEY")
-# Try different free models if one is rate-limited
-OPENROUTER_MODEL = os.getenv("OPENROUTER_MODEL", "google/gemma-2-2b-it:free")  # Alternative free model
-# Other free options: "meta-llama/llama-3.2-3b-instruct:free", "mistralai/mistral-7b-instruct:free"
+OPENROUTER_MODEL = os.getenv("OPENROUTER_MODEL", "google/gemma-2-2b-it:free")
 
 
 def get_llm() -> LLM:
@@ -42,42 +44,47 @@ def get_llm() -> LLM:
         ValueError: If required configuration is not set
     """
     if LLM_PROVIDER == "ollama":
-        # Option 1: Ollama (local open-source models)
-        # Install: brew install ollama (Mac) or see https://ollama.ai
-        # Run: ollama pull llama3.2
+        # Ollama (local models only, not for cloud deployment)
         return LLM(
             model=OLLAMA_MODEL,
             base_url=OLLAMA_BASE_URL,
-            api_key="ollama"  # Ollama doesn't require a real API key, but CrewAI needs something
+            api_key="ollama"
+        )
+    
+    elif LLM_PROVIDER == "huggingface":
+        # Hugging Face Inference API - Best open-source models
+        if not HF_API_KEY:
+            raise ValueError(
+                "HF_API_KEY environment variable is not set. "
+                "Get a free token from https://huggingface.co/settings/tokens"
+            )
+        return LLM(
+            model=f"huggingface/{HF_MODEL}",
+            api_key=HF_API_KEY
         )
     
     elif LLM_PROVIDER == "litellm":
-        # Option 2: LiteLLM (supports Hugging Face, Together AI, etc.)
-        # Format: huggingface/model-name or together_ai/model-name
+        # LiteLLM (alternative provider)
         return LLM(
             model=f"litellm/{LITELLM_MODEL}",
-            api_key=LITELLM_API_KEY if LITELLM_API_KEY else "dummy"  # Some providers don't need keys
+            api_key=LITELLM_API_KEY if LITELLM_API_KEY else "dummy"
         )
     
     elif LLM_PROVIDER == "openrouter":
-        # Option 3: OpenRouter (access to many open-source models)
-        # Free tier available: https://openrouter.ai
-        # Note: For Hugging Face Spaces, set OPENROUTER_API_KEY as a secret
+        # OpenRouter (backup option if HF is unavailable)
         if not OPENROUTER_API_KEY:
             raise ValueError(
                 "OPENROUTER_API_KEY environment variable is not set. "
                 "Get a free key at https://openrouter.ai"
             )
-        # CrewAI uses LiteLLM internally, so we need to format as openrouter/model-name
-        # Using gemma-2-2b as default (less likely to be rate-limited than llama-3.2)
         return LLM(
             model=f"openrouter/{OPENROUTER_MODEL}",
             api_key=OPENROUTER_API_KEY,
-            temperature=0.3  # Add temperature for better responses
+            temperature=0.3
         )
     
     else:
-        # Default: OpenAI (original configuration)
+        # OpenAI (paid option)
         if not OPENAI_API_KEY:
             raise ValueError(
                 "OPENAI_API_KEY environment variable is not set. "