Skip to content

HackerAIOfficial/simple-llama-finetuner

 
 

Repository files navigation

🦙 Simple LLaMA Finetuner - CLI

Open In Colab

Original README.md

Simple LLaMA Finetuner is a beginner-friendly interface designed to facilitate fine-tuning the LLaMA-7B language model using LoRA method via the PEFT library on commodity NVIDIA GPUs. With small dataset and sample lengths of 256, you can even run this on a regular Colab Tesla T4 instance.

With this intuitive UI, you can easily manage your dataset, customize parameters, train, and evaluate the model's inference capabilities.

Acknowledgements

Features

  • Simply paste datasets in the UI, separated by double blank lines
  • Adjustable parameters for fine-tuning and inference
  • Beginner-friendly UI with explanations for each parameter

TODO

  • Accelerate / DeepSpeed
  • Load other models
  • More dataset preparation tools

Getting Started

Prerequisites

  • Linux or WSL
  • Modern NVIDIA GPU with >16 GB of VRAM (but it might be possible to run with less for smaller sample lengths)

Usage

I recommend using a virtual environment to install the required packages. Conda preferred.

conda create -n llama-finetuner python=3.10
conda activate llama-finetuner
conda install -y cuda -c nvidia/label/cuda-11.7.0
conda install -y pytorch=1.13.1 pytorch-cuda=11.7 -c pytorch

On WSL, you might need to install CUDA manually by following these steps, then running the following before you launch:

export LD_LIBRARY_PATH=/usr/lib/wsl/lib

Clone the repository and install the required packages.

git clone https://github.com/lxe/simple-llama-finetuner.git
cd simple-llama-finetuner
pip install -r requirements.txt

Launch it

python main.py

Open http://127.0.0.1:7860/ in your browser. Prepare your training data by separating each sample with 2 blank lines. Paste the whole training dataset into the textbox. Specify the model name in the "LoRA Model Name" textbox, then click train. You might need to adjust the max sequence length and batch size to fit your GPU memory. The model will be saved in the lora-{your model name} directory.

After training is done, navigate to "Inference" tab, click "Reload Models", select your model, and play with it.

Have fun!

Screenshots

Image1 Image2

Cli version

Usage:

$ ./main_cli.py -h
Usage: main_cli.py command [args...]

Commands:
  predict
  train

$ ./main_cli.py train -h
Usage: main-cli.py train [OPTIONS]

Options:
  --epochs=INT                         (default: 1)
  --gradient-accumulation-steps=INT    (default: 1)
  --learning-rate=FLOAT                (default: 0.0003)
  --lora-alpha=INT                     (default: 16)
  --lora-dropout=FLOAT                 (default: 0.01)
  --lora-r=INT                         (default: 8)
  --max-seq-length=INT                 (default: 512)
  --micro-batch-size=INT               (default: 1)
  --model-name=STR                     (default: elderberry-cherry)
  --training-file=STR                  (default: ./example-datasets/leo.txt)

Other actions:
  -h, --help                          Show the help
  
$ ./main_cli.py predict -h
Usage: ./main_cli.py predict [OPTIONS]

Options:
  --inference-text=STR    (default: What is leo?)
  --max-new-tokens=INT    (default: 50)
  --model-name=STR        (default: ./lora-elderberry-cherry)
  --repeat-penalty=INT    (default: 1)
  --temperature=FLOAT     (default: 0.01)
  --top-k=INT             (default: 50)
  --top-p=FLOAT           (default: 0.3)

Other actions:
  -h, --help             Show the help


How to use?

$ ./main_cli.py train --training-file ./example-datasets/leo.txt

$ ./main_cli.py predict --inference_text "C'est qui Paulivan"
Un gros faignant

License

MIT License

Copyright (c) 2023 Aleksey Smolenchuk

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

About

Simple UI cli LLaMA Model Finetuning

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 51.4%
  • Jupyter Notebook 48.6%