standardize llm in retriever

langchain-streamlit-demo/app.py

@@ -148,6 +148,7 @@ def get_texts_and_retriever_cacheable_wrapper(
         chunk_size=chunk_size,
         chunk_overlap=chunk_overlap,
         k=k,
+        model=model,
         azure_kwargs=azure_kwargs,
         use_azure=use_azure,
     )

langchain-streamlit-demo/defaults.py

@@ -21,9 +21,7 @@ MODEL_DICT = {
 
 SUPPORTED_MODELS = list(MODEL_DICT.keys())
 
-DEFAULT_MODEL = os.environ.get(
-    "DEFAULT_MODEL", "gpt-3.5-turbo"
-)  # "gpt-4-turbo-preview")
+DEFAULT_MODEL = os.environ.get("DEFAULT_MODEL", "gpt-4-turbo-preview")
 
 DEFAULT_SYSTEM_PROMPT = os.environ.get(
     "DEFAULT_SYSTEM_PROMPT",

langchain-streamlit-demo/llm_resources.py

@@ -5,6 +5,7 @@ from typing import Dict, List, Optional, Tuple
 from defaults import (
     DEFAULT_CHUNK_OVERLAP,
     DEFAULT_CHUNK_SIZE,
+    DEFAULT_MODEL,
     DEFAULT_RETRIEVER_K,
     DEFAULT_SYSTEM_PROMPT,
 )
@@ -243,6 +244,7 @@ def get_texts_and_multiretriever(
     chunk_size: int = DEFAULT_CHUNK_SIZE,
     chunk_overlap: int = DEFAULT_CHUNK_OVERLAP,
     k: int = DEFAULT_RETRIEVER_K,
+    model: str = DEFAULT_MODEL,
     azure_kwargs: Optional[Dict[str, str]] = None,
     use_azure: bool = False,
 ) -> Tuple[List[Document], BaseRetriever]:
@@ -301,7 +303,7 @@ def get_texts_and_multiretriever(
         multiquerystore = FAISS.from_documents(multiquery_texts, embeddings)
         multiquery_retriever = MultiQueryRetriever.from_llm(
             retriever=multiquerystore.as_retriever(search_kwargs={"k": k}),
-            llm=ChatOpenAI(),
+            llm=ChatOpenAI(model=model, temperature=0.0),
         )
 
         ensemble_retriever = EnsembleRetriever(