feat(model): try mixtral-8x22b-instruct-v0.1

confluence_rag.py

@@ -87,7 +87,8 @@ def store_vector(all_splits: List[str]) -> VectorStoreRetriever:
 
     # Use the HuggingFace distilbert-base-uncased model to embed the text
     embeddings_model_url = (
-        "https://api-inference.huggingface.co/models/distilbert-base-uncased"
+        # "https://api-inference.huggingface.co/models/distilbert-base-uncased"
+        "https://api-inference.huggingface.co/models/Salesforce/SFR-Embedding-Mistral"
     )
 
     embeddings = HuggingFaceInferenceAPIEmbeddings(
@@ -138,13 +139,15 @@ def generate_rag_chain(retriever: VectorStoreRetriever = None) -> RunnableSequen
     """
     # Use the Mistral Free prototype API
     mistral_url = (
-        "https://api-inference.huggingface.co/models/mistralai/Mistral-7B-Instruct-v0.2"
+        # "https://api-inference.huggingface.co/models/mistralai/Mistral-7B-Instruct-v0.2"
+        "https://api-inference.huggingface.co/models/mistralai/Mixtral-8x22B-Instruct-v0.1"
     )
 
     model_endpoint = HuggingFaceEndpoint(
         endpoint_url=mistral_url,
         huggingfacehub_api_token=HF_API_KEY,
         task="text2text-generation",
+        max_new_tokens=1024
     )
 
     # Use a custom output parser