Merge pull request #91 from openchatai/feature/query_planner

Adding plan and executor chain

README.md

@@ -45,6 +45,12 @@ git clone [email protected]:openchatai/OpenCopilot.git
 OPENAI_API_KEY=YOUR_TOKEN_HERE
 ```
 
+- gpt-4: Ideal for more complex tasks, but may have slower processing times.
+- gpt-3.5-turbo-16k: This model is significantly faster compared to GPT-4.
+```env
+PLAN_AND_EXECUTE_MODEL=gpt-3.5-turbo-16k
+```
+
 - After updating your API key, navigate to the repository folder and run the following command (for macOS or Linux):
 
 ```

docker-compose.yml

@@ -30,6 +30,8 @@ services:
       - opencopilot_network
     env_file:
       - llm-server/.env
+    ports: 
+      - 8002:8002
     depends_on:
       - mongodb
       - qdrant

llm-server/.env.example

@@ -12,4 +12,10 @@ STORE=QDRANT
 LANGCHAIN_TRACING_V2=true
 LANGCHAIN_ENDPOINT="https://api.smith.langchain.com"
 LANGCHAIN_API_KEY="TOKEN_GOES_HERE"
-LANGCHAIN_PROJECT="PROJECT_NAME_GOES_HERE"
+LANGCHAIN_PROJECT="PROJECT_NAME_GOES_HERE"
+
+
+# use gpt-4 for more complicated tasks, but it is usually much slower than 3.5-turbo family
+# gpt-3.5-turbo-16k - this model is many times faster than gpt-4, gpt-4 is more accurate with self observation
+PLAN_AND_EXECUTE_MODEL=gpt-3.5-turbo-16k
+VECTOR_DB_THRESHOLD=0.88

llm-server/Dockerfile

@@ -15,5 +15,5 @@ COPY . /app/
 
 EXPOSE 8002
 
-# Run app.py when the container launches
-CMD ["python", "app.py"]
+# -u To prevent log accumulation, execute app.py with unbuffered output as soon as the container starts up.
+CMD ["python", "-u" ,"app.py"]