https://github.com/coffeevampir3/ez-trainer

Train Llama Loras Easily
https://github.com/coffeevampir3/ez-trainer

data-science easy-to-use llama llama2 machine-learning training webui

Last synced: about 2 months ago
JSON representation

Train Llama Loras Easily

• Host: GitHub
• URL: https://github.com/coffeevampir3/ez-trainer
• Owner: CoffeeVampir3
• License: unlicense
• Created: 2023-06-29T05:55:28.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2023-08-03T03:28:37.000Z (over 1 year ago)
• Last Synced: 2024-10-18T21:59:34.504Z (3 months ago)
• Topics: data-science, easy-to-use, llama, llama2, machine-learning, training, webui
• Language: Python
• Homepage:
• Size: 44.9 KB
• Stars: 29
• Watchers: 2
• Forks: 2
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# ez-trainer

Train Llama Loras Easily

Extremely ugly interface for easy lora training for Llama models. Put your models in the models folder, put your datasets in the datasets folder. Wow!

Datasets are expected to be json or jsonl format right now. Models should be the unquantized huggingface format. More info soon, maybe.

# Features:

### Training:

Train Loras for Llama. See wiki.

### Merging

Merge your Loras back into Llama if you want. Supports strength-based merging.

# Check out the wiki for usage info

## Install:

`pip install -r requirements.txt`

(This does not install torch, you should install that yourself. )

(Step by step instructions for exactly how to install in the wiki )

## Run:

`python train_module.py`