e2fyi / runpod-images

Dockerfile for custom runpod

runpod-images

This repo contains custom images for usage in https://www.runpod.io.

All images can be found here.

The images are of the form:

ghcr.io/e2fyi/runpod-images:<tag>

Available images

---

runpod-images:base-*:

Base pytorch image + huggingface-cli + git-lfs + pipx + poetry

The base image has a util script (/run.sh) which takes in a command argument.

# example
/run.sh python -m finetune

The script will

• first clone your hugging face model repo into /workspace/$HF_MODEL
• execute the command provided - remember to save your model weights to /workspace/$HF_MODEL
• git add/commit/push your model to hugging face
• stop the pod (NOTE NOT terminate)

You will need to provide the following env variables:

• HF_EMAIL
• HF_REPO
• HF_MODEL
• HF_TOKEN
• COMMIT_MSG

tags:

• base-v2023.3.21: py=310, pytorch=1.13.1, cuda=116, pipx=1.2, poetry=1.4, hf=0.13

---

runpod-images:alpaca-lora-*:

Project image to finetune alpaca-lora model (see https://github.com/tloen/alpaca-lora.git)

tags:

• alpaca-lora-v2023.3.21