https://allenai.github.io/clin/

https://allenai.github.io/clin/

Last synced: 3 days ago
JSON representation

• Host: GitHub
• URL: https://allenai.github.io/clin/
• Owner: allenai
• Created: 2023-10-03T19:07:27.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2023-12-15T05:30:28.000Z (about 1 year ago)
• Last Synced: 2024-04-14T07:49:56.410Z (9 months ago)
• Language: JavaScript
• Size: 105 MB
• Stars: 61
• Watchers: 6
• Forks: 8
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• awesome_ai_agents - CLIN - CLIN by Allen Institute for AI is an interactive continual learning agent that adapts rapidly to tasks, using a setup process involving Java, Python, and the ScienceWorld environment, supported by models like GPT-3.5-turbo and GPT-4 [website](https://allenai.github.io/clin/) | [github](https://github.com/allenai/clin) | [research paper](https://arxiv.org/pdf/2310.10134.pdf) (Learning / Repositories)
• awesome_ai_agents - CLIN - CLIN by Allen Institute for AI is an interactive continual learning agent that adapts rapidly to tasks, using a setup process involving Java, Python, and the ScienceWorld environment, supported by models like GPT-3.5-turbo and GPT-4 [website](https://allenai.github.io/clin/) | [github](https://github.com/allenai/clin) | [research paper](https://arxiv.org/pdf/2310.10134.pdf) (Learning / Repositories)

README

        
# CLIN: A Continually Learning Language Agent for Rapid Task Adaptation and Generalization



    

  

  

    

   

## Code for interactive continual learning with ScienceWorld environment.

Step 1: Install Java

You will have to have Java 1.8+ installed on your system (shipped with most Linux distributions).

Step 2: Create a new Python environment

```

conda create --name sw python=3.8

conda activate sw

pip install -r requirements.txt

export PYTHONPATH=.         # prevents any module errors

export OPENAI_API_KEY=  # would need access to one of these 2 models: gpt-3.5-turbo, gpt-4

```

Step 3: Installing ScienceWorld

```

cd ..

git clone https://github.com/allenai/ScienceWorld.git

cd ScienceWorld

git checkout exhaustivevalidactions

git pull

pip install -e .

```

Step 4: Get back into CLIN repo

```

cd ../clin

mkdir logs

```

Step 5: Example command to run CLIN agent for ScienceWorld

```

python scienceworld/clin_agent.py --task-num "4" --var-num 1 --env-step-limit 2 --num-episodes 1  --gpt-model "gpt-4-0613" --summarize_end_of_episode 1  --device "cpu"  --temperature 0.0  --use-gold-memory-in-ep0 0 --gold-traces "" --use-last-k-memories 3 --quadrant 1 --simplifications-preset "easy" --output-path-prefix logs/testrun/

```

## Citation

```bib

@article{majumder2023clin,

  author    = "Majumder, Bodhisattwa Prasad and Dalvi Mishra, Bhavana and Jansen, Peter and Tafjord, Oyvind and Tandon, Niket and Zhang, Li and Callison-Burch, Burch and Clark, Peter",

  title     = "CLIN: A Continually Learning Language Agent for Rapid Task Adaptation and Generalization",

  journal   = "arXiv",

  year      = "2023",

}

```