Added hf_llm_engine and config

sia/hf_llm_engine.py

@@ -0,0 +1,71 @@
+from typing import Iterator, Optional
+from huggingface_hub import InferenceClient
+
+from .llm_engine import LlmEngine
+
+class HfLlmEngine(LlmEngine):
+    """
+    LLM Engine implementation using HuggingFace's InferenceClient.
+    """
+    
+    def __init__(
+        self,
+        model_id: str = "mistralai/Mistral-7B-Instruct-v0.2",
+        api_token: Optional[str] = None,
+        temperature: float = 0.7,
+        max_new_tokens: int = 1024,
+    ):
+        """
+        Initialize the HuggingFace Inference API LLM Engine.
+
+        Args:
+            model_id: HuggingFace model ID to use (default: Mistral-7B-Instruct)
+            api_token: HuggingFace API token. If None, will try to read from HF_TOKEN env var
+            temperature: Sampling temperature (default: 0.7)
+            max_new_tokens: Maximum number of tokens to generate (default: 1024)
+        """
+        self.model_id = model_id
+        self.client = InferenceClient(token=api_token)
+        
+        # Generation parameters
+        self.temperature = temperature
+        self.max_new_tokens = max_new_tokens
+            
+    def set_model_path(self, model_id: str):
+        """
+        Update the model being used.
+
+        Args:
+            model_id: New HuggingFace model ID to use
+        """
+        self.model_id = model_id
+                        
+    def infer(self, system_prompt: str, main_context: str) -> Iterator[str]:
+        """
+        Run inference using the system prompt and main context.
+
+        Args:
+            system_prompt: The system prompt string
+            main_context: The main context string after templating
+
+        Returns:
+            Iterator[str]: An iterator that yields the generated text.
+        """
+        messages = [
+            {"role": "system", "content": system_prompt},
+            {"role": "user", "content": main_context}
+        ]
+        
+        def stream_wrapper():
+            stream = self.client.chat_completion(
+                model=self.model_id,
+                messages=messages,
+                temperature=self.temperature,
+                max_tokens=self.max_new_tokens,
+                stream=True
+            )
+            
+            for response in stream:
+                if content := response.choices[0].delta.content:
+                    yield content
+        return stream_wrapper()