Added Dockerfile

[t46]

I've added a Dockerfile. However, there's no specific need for this file - I just thought having a Dockerfile might be convenient.Since installing texlive took a long time for me, I used a texlive-based image for now, but I'm not sure if that's the best approach. Also, I'm not certain if the RUN commands I included are desirable. If there's a better way to write the file, please feel free to use that instead.

I've also created a docker image for reference:
https://hub.docker.com/repository/docker/t4646/ai-scientist/general

NOTE: I uploaded the newer version

[t46]

I updated the Dorckerfile. This is now dependent on python:3.11-bullseye image.

I also uploaded the docker image based on this file. This image do all in readme except for nanogpt&lite setup and api key setting. so you can run
python launch_scientist.py --model "gpt-4o-2024-05-13" --experiment 2d_diffusion --num-ideas 2
for example.

[JGalego]

Just a few recommendations from an outsider:

• Check your Dockerfile against a good linter e.g. Hadolint
• Consider adding an entrypoint script for the image so that running launch_scientist.py is as easy as

  docker run sakana-ai/ai-scientist \
     -e OPENAI_API_KEY=$OPENAI_API_KEY \
     --model "gpt-4o-2024-05-13" \
     --experiment 2d_diffusion \
     --num-ideas 2

• A word to the owners: how can we bring the image size down (currently at ~6GB)? Is texlive-full really necessary?

[t46]

I've largely incorporated the points @JGalego pointed out, except for using texlive-full and ignoring some of the linter's suggestions (using cd in a loop).

• Check the Dockerfile against Hadolint
• Add entrypoint script to run launch_scientist.py

The image is here:
https://hub.docker.com/layers/t4646/ai-scientist/20240817/images/sha256-1a10dccad33cbf1f25f07cd34f080d37fe6e16eb0933fd3915cf7f40116a73bf?context=repo

You can use this image like this:
[endpoint script]

docker run -e OPENAI_API_KEY=$OPENAI_API_KEY t4646/ai-scientist:20240817 \
       --model “gpt-4o-2024-05-13” \
       --experiment 2d_diffusion \
       --num-ideas 1

[interactive]

docker run -it -e OPENAI_API_KEY=$OPENAI_API_KEY \
  --entrypoint /bin/bash \
  t4646/ai-scientist:20240817

[curiosityz]

TeLive Full eats up all the space when trying to run this on Google Project IdX an Github Codespaces

[conglu1997]

Thank you for your contribution! Indeed, something more lightweight than texlive-full would be excellent and something that would be greatly appreciated!

[curiosityz]

I keep getting NVIDIA CUDA errors what I need to install, I get it on all the places I've tried to install also locally

[conglu1997]

What version is your CUDA driver?

[dmvieira]

If you want to save your running results need this command:

docker run -e OPENAI_API_KEY=$OPENAI_API_KEY -v `pwd`/templates:/app/AI-Scientist/templates t4646/ai-scientist:20240817 \
       --model gpt-4o-2024-05-13 \
       --experiment 2d_diffusion \
       --num-ideas 1

It will be inside templates/2d_diffusion/run_* folder

[conglu1997]

Thanks, changed! :)

[callor]

Thank you very much
I tried to run it locally in a windows environment, but it didn't work, so I'm running it in a docker environment.
There are some errors during the execution process, but
I'm curious about how to set the API key for https://www.semanticscholar.org/

[callor]

In interactive environment, I still get CUDA Disable related errors during model training. I don't know why.

[aaronmandell]

Any folks interested/willing to help integrate AIScientist into a DeSci DAO, please DM.

[callor]

Hello Dear
During the Docker execution process, it keeps stopping, leaving only the following log. What could be the problem?

System : Windows 11 x64
GPU : RTX A2000
Memory : 32GB

Please Help Me!!

shell command

docker run -d --gpus all --env-file=/app/ai-project/.env -v "/$(pwd)/templates:/app/ai-project/AI-Scientist/templa
tes" t4646/ai-scientist:20240817 --model gpt-4o-2024-05-13 --experiment nanoGPT --num-ideas 2

log message

2024-09-09 07:57:20 Response Status Code: 200
2024-09-09 07:57:20 Response Content: {"total": 12, "offset": 0, "next": 10, "data": [{"paperId": "910aea4a020c329afb8b8a948abaafdf9ebcab3f", "title": "DropAttention: A Regularization Method for Fully-Connected Self-Attention Networks", "abstract": "Variants dropout methods have been designed for the fully-connected layer, convolutional layer and recurrent layer in neural networks, and shown to be effective to avoid overfitting. As an appealing alternative to recurrent and convolutional layers, the fully-connected self-attention lay
2024-09-09 07:57:20 Decision made: novel after round 3
2024-09-09 07:57:20 
2024-09-09 07:57:20 Checking novelty of idea 3: adaptive_learning_rate
2024-09-09 07:57:20 Response Status Code: 200
2024-09-09 07:57:20 Response Content: {"total": 707, "offset": 0, "next": 10, "data": [{"paperId": "76488b0743c9553b7b1d7ec46afe107ea60a67ca", "title": "AutoLRS: Automatic Learning-Rate Schedule by Bayesian Optimization on the Fly", "abstract": "The learning rate (LR) schedule is one of the most important hyper-parameters needing careful tuning in training DNNs. However, it is also one of the least automated parts of machine learning systems and usually costs significant manual effort and computing. Though there are pre-defined LR s
2024-09-09 07:57:20 Response Status Code: 200
2024-09-09 07:57:20 Response Content: {"total": 53, "offset": 0, "next": 10, "data": [{"paperId": "8508a2660777b2cf9130530f690f58896b61bdf8", "title": "A Cascade CNN Model based on Adaptive Learning Rate Thresholding for Reliable Face Recognition", "abstract": "Convolutional models may effectively identify persons quickly through automatic face analysis. A convolutional neural network architecture known as the CNN cascade structure uses numerous deep convolution layers to extract hierarchical characteristics from the input image. Ca
2024-09-09 07:57:20 Decision made: novel after round 2
2024-09-09 07:57:20 Processing idea: adaptive_block_size
2024-09-09 07:57:20 Failed to evaluate idea adaptive_block_size: [Errno 2] No such file or directory: 'templates/nanoGPT/run_0/final_info.json'
2024-09-09 07:57:20 Processing idea: attention_dropout
2024-09-09 07:57:20 Failed to evaluate idea attention_dropout: [Errno 2] No such file or directory: 'templates/nanoGPT/run_0/final_info.json'
2024-09-09 07:57:20 Processing idea: adaptive_learning_rate
2024-09-09 07:57:20 Failed to evaluate idea adaptive_learning_rate: [Errno 2] No such file or directory: 'templates/nanoGPT/run_0/final_info.json'
2024-09-09 07:57:20 All ideas evaluated.

[curiosityz]

Well look at the error....

2024-09-09 07:57:20 Failed to evaluate idea adaptive_block_size: [Errno 2] No such file or directory: 'templates/nanoGPT/run_0/final_info.json'

It sounds like they didn't build the pre-run steps into Docker. Go read the directions and run the steps it says "you need to run this first"