Merge pull request #152 from nulib/wip

Turn a POSTed question into a valid JSON response from the LLM

python/src/handlers/chat.py

@@ -2,48 +2,79 @@
 import json
 import os
 import setup
+from langchain.chains import RetrievalQAWithSourcesChain
 
-def handler(event, context):
-  question = event.get("body", "")
-  if event.get("isBase64Encoded", False):
-    question = base64.b64decode(question)
 
-  headers = event.get("headers")
-  token = headers.get("authorization", headers.get("Authorization", None))
 
-  if token is None:
-    for cookie in event.get("cookies", []):
-      [k, v] = cookie.split("=", 1)
-      if k == os.getenv("API_TOKEN_NAME"):
-        token = v
-  else:
-    token = token.replace("Bearer ", "")
 
-  if not setup.validate_token(token):
+def handler(event, context):
+  if not is_authenticated(event):
     return {
     "statusCode": 401,
     "headers": {
       "Content-Type": "text/plain"
     },
     "body": "Unauthorized"
   }
-
-
-  params = event.get("queryStringParameters", {})
-  index_name = params.get("index", "Work")
-  text_key = params.get("text_key", "title")
-  attributes = params.get("attributes", "identifier,title").split(",")
+  question = get_query(event)
+  index_name = get_param(event, "index", "Work")
+  text_key = get_param(event, "text_key", "title")  
+  attributes = get_param(event, 
+                         "attributes", 
+                         "identifier,title,source,alternate_title,contributor,create_date,creator,date_created,description,genre,keywords,language,location,physical_description_material,physical_description_size,scope_and_contents,style_period,subject,table_of_contents,technique,work_type").split(",")
 
   weaviate = setup.weaviate_vector_store(index_name=index_name, 
                                          text_key=text_key, 
                                          attributes=attributes)
-  result = weaviate.similarity_search_by_text(query=question, 
-                                              additional="certainty")
+
+  client = setup.openai_chat_client()
 
+
+  chain = RetrievalQAWithSourcesChain.from_chain_type(
+    client, 
+    chain_type="stuff", 
+    retriever=weaviate.as_retriever(),
+    return_source_documents=True)
+
+  response = chain({"question": question})
+  print(response)
+  response['source_documents'] = [doc.__dict__ for doc in response['source_documents']]
   return {
     "statusCode": 200,
     "headers": {
       "Content-Type": "application/json"
     },
-    "body": json.dumps([doc.__dict__ for doc in result])
-  }
+    "body": json.dumps(response)
+  }
+
+
+def get_param(event, parameter, default):
+  params = event.get("queryStringParameters", {})
+  return params.get(parameter, default)
+
+
+def get_query(event):
+  question = event.get("body", "")
+  if event.get("isBase64Encoded", False):
+    question = base64.b64decode(question)
+  return question
+
+
+def is_authenticated(event):
+  headers = event.get("headers")
+  token = headers.get("authorization", headers.get("Authorization", None))
+
+  if token is None:
+    for cookie in event.get("cookies", []):
+      [k, v] = cookie.split("=", 1)
+      if k == os.getenv("API_TOKEN_NAME"):
+        token = v
+  else:
+    token = token.replace("Bearer ", "")
+
+  return setup.validate_token(token)
+
+
+
+# result = weaviate.similarity_search_by_text(query=question, 
+#                                             additional="certainty")

python/src/setup.py

@@ -7,24 +7,25 @@
 
 def openai_chat_client():
   deployment = os.getenv("AZURE_OPENAI_LLM_DEPLOYMENT_ID")
-  os.getenv("AZURE_OPENAI_RESOURCE_NAME")
-  os.getenv("AZURE_OPENAI_API_KEY")
+  key = os.getenv("AZURE_OPENAI_API_KEY")
+  resource = os.getenv("AZURE_OPENAI_RESOURCE_NAME")
+
+  return AzureChatOpenAI(deployment_name=deployment, 
+                         openai_api_key=key, 
+                         openai_api_base=f"https://{resource}.openai.azure.com/",
+                         openai_api_version="2023-03-15-preview")
 
-  return AzureChatOpenAI(deployment_name=deployment)
 
 def weaviate_vector_store(index_name: str, text_key: str, attributes: List[str] = []):
   weaviate_url = os.environ['WEAVIATE_URL']
   weaviate_api_key = os.environ['WEAVIATE_API_KEY']
-  openai_api_key = os.environ['AZURE_OPENAI_API_KEY']
+  # openai_api_key = os.environ['AZURE_OPENAI_API_KEY']
 
   auth_config = weaviate.AuthApiKey(api_key=weaviate_api_key)
 
   client = weaviate.Client(
       url=weaviate_url,
-      auth_client_secret=auth_config,
-      additional_headers={
-          "X-OpenAI-Api-Key": openai_api_key
-      }
+      auth_client_secret=auth_config
   )
   return Weaviate(client=client, 
                   index_name=index_name, 

template.yaml

@@ -554,7 +554,7 @@ Resources:
       CodeUri: ./python/src
       Runtime: python3.9
       Handler: handlers/chat.handler
-      Timeout: 30
+      Timeout: 300
       Environment:
         Variables:
           AZURE_OPENAI_API_KEY: !Ref AzureOpenaiApiKey