Merge pull request #269 from larinam/main

Azure implementation for the streaming output

README.md

@@ -64,7 +64,7 @@ Note: Make sure you have docker installed
    OPENAI_API_KEY=Yourkey
    VITE_API_STREAMING=true
    ```
-3. Run `docker-compose build && docker-compose up`
+3. Run `./run-with-docker-compose.sh`
 4. Navigate to http://localhost:5173/
 
 To stop just run Ctrl + C

application/.env_sample

@@ -8,4 +8,5 @@ API_URL=http://localhost:5001
 #For OPENAI on Azure
 OPENAI_API_BASE=
 OPENAI_API_VERSION=
-AZURE_DEPLOYMENT_NAME=
+AZURE_DEPLOYMENT_NAME=
+AZURE_EMBEDDINGS_DEPLOYMENT_NAME=

application/app.py

@@ -2,7 +2,9 @@ import asyncio
 import datetime
 import http.client
 import json
+import logging
 import os
+import platform
 import traceback
 
 import dotenv
@@ -40,6 +42,8 @@ from worker import ingest_worker
 
 # os.environ["LANGCHAIN_HANDLER"] = "langchain"
 
+logger = logging.getLogger(__name__)
+
 if settings.LLM_NAME == "manifest":
     from manifest import Manifest
     from langchain.llms.manifest import ManifestWrapper
@@ -47,7 +51,6 @@ if settings.LLM_NAME == "manifest":
     manifest = Manifest(client_name="huggingface", client_connection="http://127.0.0.1:5000")
 
 # Redirect PosixPath to WindowsPath on Windows
-import platform
 
 if platform.system() == "Windows":
     import pathlib
@@ -124,7 +127,12 @@ def get_vectorstore(data):
 
 def get_docsearch(vectorstore, embeddings_key):
     if settings.EMBEDDINGS_NAME == "openai_text-embedding-ada-002":
-        docsearch = FAISS.load_local(vectorstore, OpenAIEmbeddings(openai_api_key=embeddings_key))
+        if is_azure_configured():
+            os.environ["OPENAI_API_TYPE"] = "azure"
+            openai_embeddings = OpenAIEmbeddings(model=settings.AZURE_EMBEDDINGS_DEPLOYMENT_NAME)
+        else:
+            openai_embeddings = OpenAIEmbeddings(openai_api_key=embeddings_key)
+        docsearch = FAISS.load_local(vectorstore, openai_embeddings)
     elif settings.EMBEDDINGS_NAME == "huggingface_sentence-transformers/all-mpnet-base-v2":
         docsearch = FAISS.load_local(vectorstore, HuggingFaceHubEmbeddings())
     elif settings.EMBEDDINGS_NAME == "huggingface_hkunlp/instructor-large":
@@ -149,7 +157,20 @@ def home():
 
 def complete_stream(question, docsearch, chat_history, api_key):
     openai.api_key = api_key
-    llm = ChatOpenAI(openai_api_key=api_key)
+    if is_azure_configured():
+        logger.debug("in Azure")
+        openai.api_type = "azure"
+        openai.api_version = settings.OPENAI_API_VERSION
+        openai.api_base = settings.OPENAI_API_BASE
+        llm = AzureChatOpenAI(
+            openai_api_key=api_key,
+            openai_api_base=settings.OPENAI_API_BASE,
+            openai_api_version=settings.OPENAI_API_VERSION,
+            deployment_name=settings.AZURE_DEPLOYMENT_NAME,
+        )
+    else:
+        logger.debug("plain OpenAI")
+        llm = ChatOpenAI(openai_api_key=api_key)
     docs = docsearch.similarity_search(question, k=2)
     # join all page_content together with a newline
     docs_together = "\n".join([doc.page_content for doc in docs])
@@ -174,9 +195,9 @@ def complete_stream(question, docsearch, chat_history, api_key):
                     messages_combine.append({"role": "user", "content": i["prompt"]})
                     messages_combine.append({"role": "system", "content": i["response"]})
     messages_combine.append({"role": "user", "content": question})
-    completion = openai.ChatCompletion.create(model="gpt-3.5-turbo",
+    completion = openai.ChatCompletion.create(model="gpt-3.5-turbo", engine=settings.AZURE_DEPLOYMENT_NAME,
                                               messages=messages_combine, stream=True, max_tokens=500, temperature=0)
-    
+
     for line in completion:
         if "content" in line["choices"][0]["delta"]:
             # check if the delta contains content
@@ -217,6 +238,10 @@ def stream():
     )
 
 
+def is_azure_configured():
+    return settings.OPENAI_API_BASE and settings.OPENAI_API_VERSION and settings.AZURE_DEPLOYMENT_NAME
+
+
 @app.route("/api/answer", methods=["POST"])
 def api_answer():
     data = request.get_json()
@@ -244,7 +269,8 @@ def api_answer():
             input_variables=["context", "question"], template=template_quest, template_format="jinja2"
         )
         if settings.LLM_NAME == "openai_chat":
-            if settings.OPENAI_API_BASE and settings.OPENAI_API_VERSION and settings.AZURE_DEPLOYMENT_NAME:  # azure
+            if is_azure_configured():
+                logger.debug("in Azure")
                 llm = AzureChatOpenAI(
                     openai_api_key=api_key,
                     openai_api_base=settings.OPENAI_API_BASE,
@@ -252,6 +278,7 @@ def api_answer():
                     deployment_name=settings.AZURE_DEPLOYMENT_NAME,
                 )
             else:
+                logger.debug("plain OpenAI")
                 llm = ChatOpenAI(openai_api_key=api_key)  # optional parameter: model_name="gpt-4"
             messages_combine = [SystemMessagePromptTemplate.from_template(chat_combine_template)]
             if history:

application/core/settings.py

@@ -18,7 +18,8 @@ class Settings(BaseSettings):
     EMBEDDINGS_KEY: str = None  # api key for embeddings (if using openai, just copy API_KEY
     OPENAI_API_BASE: str = None  # azure openai api base url
     OPENAI_API_VERSION: str = None  # azure openai api version
-    AZURE_DEPLOYMENT_NAME: str = None  # azure deployment name
+    AZURE_DEPLOYMENT_NAME: str = None  # azure deployment name for answering
+    AZURE_EMBEDDINGS_DEPLOYMENT_NAME: str = None  # azure deployment name for embeddings
 
 
 path = Path(__file__).parent.parent.absolute()

application/worker.py

@@ -19,9 +19,11 @@ try:
 except FileExistsError:
     pass
 
+
 def metadata_from_filename(title):
     return {'title': title}
 
+
 def generate_random_string(length):
     return ''.join([string.ascii_letters[i % 52] for i in range(length)])
 

docker-compose-azure.yaml

@@ -0,0 +1,71 @@
+version: "3.9"
+
+services:
+  frontend:
+    build: ./frontend
+    environment:
+      - VITE_API_HOST=http://localhost:5001
+      - VITE_API_STREAMING=$VITE_API_STREAMING
+    ports:
+      - "5173:5173"
+    depends_on:
+      - backend
+
+  backend:
+    build: ./application
+    environment:
+      - API_KEY=$OPENAI_API_KEY
+      - EMBEDDINGS_KEY=$OPENAI_API_KEY
+      - CELERY_BROKER_URL=redis://redis:6379/0
+      - CELERY_RESULT_BACKEND=redis://redis:6379/1
+      - MONGO_URI=mongodb://mongo:27017/docsgpt
+      - OPENAI_API_KEY=$OPENAI_API_KEY
+      - OPENAI_API_BASE=$OPENAI_API_BASE
+      - OPENAI_API_VERSION=$OPENAI_API_VERSION
+      - AZURE_DEPLOYMENT_NAME=$AZURE_DEPLOYMENT_NAME
+      - AZURE_EMBEDDINGS_DEPLOYMENT_NAME=$AZURE_EMBEDDINGS_DEPLOYMENT_NAME
+    ports:
+      - "5001:5001"
+    volumes:
+      - ./application/indexes:/app/indexes
+      - ./application/inputs:/app/inputs
+      - ./application/vectors:/app/vectors
+    depends_on:
+        - redis
+        - mongo
+
+  worker:
+    build: ./application
+    command: celery -A app.celery worker -l INFO
+    environment:
+      - API_KEY=$OPENAI_API_KEY
+      - EMBEDDINGS_KEY=$OPENAI_API_KEY
+      - CELERY_BROKER_URL=redis://redis:6379/0
+      - CELERY_RESULT_BACKEND=redis://redis:6379/1
+      - MONGO_URI=mongodb://mongo:27017/docsgpt
+      - API_URL=http://backend:5001
+      - OPENAI_API_KEY=$OPENAI_API_KEY
+      - OPENAI_API_BASE=$OPENAI_API_BASE
+      - OPENAI_API_VERSION=$OPENAI_API_VERSION
+      - AZURE_DEPLOYMENT_NAME=$AZURE_DEPLOYMENT_NAME
+      - AZURE_EMBEDDINGS_DEPLOYMENT_NAME=$AZURE_EMBEDDINGS_DEPLOYMENT_NAME
+    depends_on:
+        - redis
+        - mongo
+
+  redis:
+    image: redis:6-alpine
+    ports:
+      - 6379:6379
+
+  mongo:
+    image: mongo:6
+    ports:
+      - 27017:27017
+    volumes:
+      - mongodb_data_container:/data/db
+
+
+
+volumes:
+  mongodb_data_container:

docker-compose.yaml

@@ -19,9 +19,6 @@ services:
       - CELERY_BROKER_URL=redis://redis:6379/0
       - CELERY_RESULT_BACKEND=redis://redis:6379/1
       - MONGO_URI=mongodb://mongo:27017/docsgpt
-      #- OPENAI_API_BASE=$OPENAI_API_BASE
-      #- OPENAI_API_VERSION=$OPENAI_API_VERSION
-      #- AZURE_DEPLOYMENT_NAME=$AZURE_DEPLOYMENT_NAME
     ports:
       - "5001:5001"
     volumes:
@@ -42,9 +39,6 @@ services:
       - CELERY_RESULT_BACKEND=redis://redis:6379/1
       - MONGO_URI=mongodb://mongo:27017/docsgpt
       - API_URL=http://backend:5001
-      #- OPENAI_API_BASE=$OPENAI_API_BASE
-      #- OPENAI_API_VERSION=$OPENAI_API_VERSION
-      #- AZURE_DEPLOYMENT_NAME=$AZURE_DEPLOYMENT_NAME
     depends_on:
         - redis
         - mongo

run-with-docker-compose.sh

@@ -0,0 +1,11 @@
+#!/bin/bash
+
+source .env
+
+if [[ -n "$OPENAI_API_BASE" ]] && [[ -n "$OPENAI_API_VERSION" ]] && [[ -n "$AZURE_DEPLOYMENT_NAME" ]] && [[ -n "$AZURE_EMBEDDINGS_DEPLOYMENT_NAME" ]]; then
+  echo "Running Azure Configuration"
+  docker-compose -f docker-compose-azure.yaml build && docker-compose -f docker-compose-azure.yaml up
+else
+  echo "Running Plain Configuration"
+  docker-compose build && docker-compose up
+fi

scripts/parser/open_ai_func.py

@@ -49,6 +49,7 @@ def call_openai_api(docs, folder_name):
         os.environ.get("OPENAI_API_BASE")
         and os.environ.get("OPENAI_API_VERSION")
         and os.environ.get("AZURE_DEPLOYMENT_NAME")
+        and os.environ.get("AZURE_EMBEDDINGS_DEPLOYMENT_NAME")
     ):
         os.environ["OPENAI_API_TYPE"] = "azure"
         openai_embeddings = OpenAIEmbeddings(model=os.environ.get("AZURE_EMBEDDINGS_DEPLOYMENT_NAME"))