Skip to content

Latest commit

 

History

History
247 lines (177 loc) · 7.69 KB

install.md

File metadata and controls

247 lines (177 loc) · 7.69 KB
Raw
myst
html_meta
title description keywords
AutoRAG - Installation and Setup
Learn how to install AutoRAG
AutoRAG,RAG,AutoRAG install

Installation and Setup

To install AutoRAG, you can use pip:

pip install AutoRAG

Plus, it is recommended to install PyOpenSSL and nltk libraries for full features.

pip install --upgrade pyOpenSSL
pip install nltk
python3 -c "import nltk; nltk.download('punkt_tab')"
python3 -c "import nltk; nltk.download('averaged_perceptron_tagger_eng')"
Do you have any trouble with installation?
First, you can check out the [troubleshooting](troubleshooting.md) page.

Note for Windows Users

AutoRAG is not fully supported on Windows yet. There are several constraints for Windows users.

  1. TART, UPR, and MonoT5 passage rerankers does not support Windows.
  2. Parsing might be not working properly in the Windows environment.
  3. Cannot use FlagEmbedding passage reranker with batch setting with 1. The default batch is 64.

Due to the constraints, we recommend using Docker images for running AutoRAG on Windows.

Plus, you MAKE SURE UPGRADE UP TO v0.3.1 for Windows users.

Installation for Local Models 🏠

For using local models, you need to install some additional dependencies.

pip install "AutoRAG[gpu]"

Installation for Parsing 🌲

For parsing you need to install some local packages like libmagic, tesseract, and poppler. The installation method depends upon your OS.

After installing this, you can install AutoRAG with parsing like below.

pip install "AutoRAG[parse]"

Installation for Korean 🇰🇷

You can install optional dependencies for the Korean language.

pip install "AutoRAG[ko]"

And after that, you have to install jdk 17 for using konlpy. Plus, remember to set environment PATH for jdk. (JAVA_HOME and PATH)

The instruction for Mac users is here.

Installation for Japanese 🇯🇵

pip install "AutoRAG[ja]"

Setup OPENAI API KEY

To use LLM and embedding models, it is common to use OpenAI models. If you want to use other models, check out here

You need to set OPENAI_API_KEY environment variable. You can get your API key at here.

export OPENAI_API_KEY="sk-...your-api-key..."

Or, as an alternative, you can set env variable using the .env file.

pip install python-dotenv

Then, make an.env file at your root folder like below.

OPENAI_API_KEY=sk-...your-api-key...

And when you try to run AutoRAG, you can use below code to load .env file.

from dotenv import load_dotenv

load_dotenv()

And you are ready to use AutoRAG!

Build from source

If you want to build AutoRAG from source, the first step is to clone the AutoRAG repository.

git clone https://github.com/Marker-Inc-Korea/AutoRAG.git

And install AutoRAG to editable.

cd AutoRAG
pip install -e .

And then, for testing and documentation build, you need to install some additional packages.

pip install -r tests/requirements.txt
pip install -r docs/requirements.txt

For testing, you have to set the environment variable at pytest.ini. Make a new pytest.ini file at the root of the project and write below.

[pytest]
env =
    OPENAI_API_KEY=sk-...your-api-key...

log_cli=true
log_cli_level=INFO

After that, you can run tests with pytest.

python -m pytest -n auto

After this, please check out our documentation for contributors. We are writing this documentation for contributors, so please wait for a while.

Run AutoRAG with 🐳 Docker

Tip: If you want to build an image for a gpu version, you can use autoraghq/autorag:gpu or autoraghq/autorag:gpu-parsing

To run AutoRAG using Docker, follow these steps:

1. Build the Docker Image

docker build --target production -t autorag:prod .

This command will build the production-ready Docker image, using only the production stage defined in the Dockerfile.

2. Run the Docker Container

Run the container with the following command:

docker run --rm -it \
  -v ~/.cache/huggingface:/root/.cache/huggingface \
  -v $(pwd)/sample_config:/usr/src/app/sample_config \
  -v $(pwd)/projects:/usr/src/app/projects \
  autoraghq/autorag:api evaluate \
  --config /usr/src/app/sample_config/rag/simple/simple_openai.yaml \
  --qa_data_path /usr/src/app/projects/test01/qa_validation.parquet \
  --corpus_data_path /usr/src/app/projects/test01/corpus.parquet \
  --project_dir /usr/src/app/projects/test01

Explanation:

  • -v ~/.cache/huggingface:/root/.cache/huggingface: Mounts the host's Hugging Face cache to the container, allowing it to access pre-downloaded models.
  • -v $(pwd)/sample_config:/usr/src/app/sample_config: Mounts the local sample_config directory to the container.
  • -v $(pwd)/projects:/usr/src/app/projects: Mounts the local projects directory to the container.
  • autoraghq/autorag:all evaluate: Executes the evaluate command inside the autoraghq/autorag:all container.
  • --config, --qa_data_path, --corpus_data_path, --project_dir: Specifies paths to the configuration file, QA dataset, corpus data, and project directory.

3. Using a Custom Cache Directory with HF_HOME

Alternatively, you can mount the Hugging Face cache to a custom location inside the container and set the HF_HOME environment variable:

docker run --rm -it \
  -v ~/.cache/huggingface:/cache/huggingface \
  -v $(pwd)/sample_config:/usr/src/app/sample_config \
  -v $(pwd)/projects:/usr/src/app/projects \
  -e HF_HOME=/cache/huggingface \
  autoraghq/autorag:api evaluate \
  --config /usr/src/app/sample_config/rag/simple/simple_openai.yaml \
  --qa_data_path /usr/src/app/projects/test01/qa_validation.parquet \
  --corpus_data_path /usr/src/app/projects/test01/corpus.parquet \
  --project_dir /usr/src/app/projects/test01

Explanation:

  • -v ~/.cache/huggingface:/cache/huggingface: Mounts the host's Hugging Face cache to /cache/huggingface inside the container.
  • -e HF_HOME=/cache/huggingface: Sets the HF_HOME environment variable to point to the mounted cache directory.

5. Debugging and Manual Access

To manually access the container for debugging or testing, start a Bash shell:

docker run --rm -it --entrypoint /bin/bash autoraghq/autorag:api

This command allows you to explore the container’s filesystem, run commands manually, or inspect logs for troubleshooting.

6. Use gpu version

To use the gpu version, you must install CUDA and cuDNN in your host system. It built on the cuda 11.8 version and pytorch docker image.

docker run --rm -it \
  -v ~/.cache/huggingface:/cache/huggingface \
  -v $(pwd)/sample_config:/usr/src/app/sample_config \
  -v $(pwd)/projects:/usr/src/app/projects \
  -e HF_HOME=/cache/huggingface \
  --gpus all \ # Be sure to add this line
  autoraghq/autorag:gpu evaluate \
  --config /usr/src/app/sample_config/rag/simple/simple_openai.yaml \
  --qa_data_path /usr/src/app/projects/test01/qa_validation.parquet \
  --corpus_data_path /usr/src/app/projects/test01/corpus.parquet \
  --project_dir /usr/src/app/projects/test01

Additional Notes

  • Ensure that the necessary directories (sample_config and projects) are present in the host system.
  • If running in a CI/CD pipeline, consider using environment variables or .env files to manage API keys and paths dynamically.