OpenAI Tune

A pleasant CLI tool for fine-tuning OpenAI models

Quickstart

1. Confirm you have Go (version>=1.16) installed:

go version

2. Install openai-tune:

go install github.com/rossheat/openai-tune@latest

3. Confirm installation:

openai-tune

If you see "command not found" you may need to add $GOPATH/bin to your PATH:

For bash:

echo 'export PATH=$PATH:$(go env GOPATH)/bin' >> ~/.bashrc && source ~/.bashrc

For zsh:

echo 'export PATH=$PATH:$(go env GOPATH)/bin' >> ~/.zshrc && source ~/.zshrc

4. Add your OpenAI API key to your environment:

export OPENAI_API_KEY=<OPENAI_API_KEY>

Usage

Upload a JSONL file for fine-tuning:

openai-tune upload -file <path-to-jsonl-file>

List all uploaded files with the purpose 'fine-tune':

openai-tune upload -list

Create a fine-tuning job with default settings:

openai-tune create -file-id <file-id> -model <model-name>

See this list of models that can be fine-tuned.

Create a fine-tuning job with custom settings from a YAML config:

openai-tune create -config <path-to-yaml>

See config.example.yml for all available options.

List fine-tuning jobs:

openai-tune list [-limit <n>] [-after <job-id>]

Get information about a specific fine-tuning job:

openai-tune get <job-id>

Cancel a fine-tuning job:

openai-tune cancel <job-id>

Weights & Biases Integration

In order to enable logging with W&B, first please follow these instructions to enable the W&B integration for your OpenAI account.

Once you've enabled the W&B integration on your OpenAI account, specified your W&B integration in your config file, and created your fine-tuning job, you can view the job in W&B by navigating to the W&B project you specified in the job creation request. Your run should be located at the URL: https://wandb.ai/<WANDB-ENTITY>/<WANDB-PROJECT>/runs/ftjob-ABCDEF