FinRobot RAG API - Docker Deployment

This README provides instructions for running the FinRobot RAG API using Docker.

Prerequisites

• Docker and Docker Compose installed
• OpenAI API key (create OAI_CONFIG_LIST file based on the OAI_CONFIG_LIST.local sample)
• API keys for financial data sources (create config_api_keys file based on the config_api_keys.local sample)
• Report PDF files (placed in the report directory)

Required Files

Before building and running the Docker container, ensure you have:

1. OAI_CONFIG_LIST file in the root directory with your OpenAI API key(s)
2. PDF files in the report directory:
   • Microsoft_Annual_Report_2023.pdf
   • 2023-07-27_10-K_msft-20230630.htm.pdf

Building and Running

# Build docker (need to modify $IMAGE_NAME:$IMAGE_TAG for example finagents:latest)
docker image build \
  --build-arg username=$USER \
  --build-arg uid=$UID \
  --build-arg gid=$GID \
  --file Dockerfile \
  --tag finagents:latest \
  ./

# Run container expose api (with coressponding $IMAGE_NAME:$IMAGE_TAG for example finagents:latest)
docker run -p 8888:8888 finagentsapi:latest                       

# View logs
docker-compose logs -f

# Stop the containers
docker-compose down

# Rebuild if you make changes
docker-compose up -d --build

Can run in local

Set up environment

Make sure you are in main directory

Conda

# creae conda environment
conda env create -f my_conda.yml

# activate conda environment
conda activate finrobot-test

Python package

pip install requirements.txt

Run fastapi server to test agents api

Currently main_up.py and main.py is stable for running fastapi request, I still developing a python file which utilize agents power of this repo. Here is the following sample for run in local

# run main_up.py
uvicorn main_up:app

Once running, the following endpoints are available:

main_up.py

• {base_url}/ - Health check endpoint
• {base_url}/chat_rag_up - Main RAG API endpoint (POST)

main.py

• {base_url}/ - Health check endpoint
• {base_url}/chat_rag - Main RAG API endpoint (POST) for querying annual reports and 10-K filings

Example curl request

main_up.py (however this file have hardcord file uploaded so user can only question and answer that only file, which I need to develop a separate api endpoint but this is for testing the main endpoing)

curl -X POST "{base_url}/chat_rag" \
  -H "Content-Type: application/json" \
  -d '{"query": "What is the strategy of Nvidia for artificial intelligence?"}'

main.py

curl -X POST "{base_url}/chat_rag_up" \
  -H "Content-Type: application/json" \
  -d '{"query": "What is the strategy of Nvidia for artificial intelligence?"}'

Persistent Data

The Docker setup includes persistent volumes for the vector databases:

• finrobot-earnings-db - Earnings call database
• finrobot-sec-db - SEC filings database
• finrobot-sec-md-db - SEC markdown filings database

Current Issue

• Still have conflict in building docker process
• install too many requiremnts.txt due to no exact dependencies cost a lot of time
• still cannot expose api

ISSUE resolution

In the running docker process if you found this bug:

ImportError: /home/zap/miniconda3/lib/python3.12/site-packages/sklearn/utils/../../scikit_learn.libs/libgomp-d22c30c5.so.1.0.0: cannot allocate memory in static TLS block

you can just copy that path expose via ENV LD_PRELOAD (follow this link for more detail keras-team/keras-tuner#317 (comment))