title | emoji | colorFrom | colorTo | sdk | app_file | pinned |
---|---|---|---|---|---|---|
Simple LLM Finetuner |
🦙 |
yellow |
orange |
gradio |
app.py |
false |
Simple LLM Finetuner is a beginner-friendly interface designed to facilitate fine-tuning various language models 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.
- https://github.com/zphang/minimal-llama/
- https://github.com/tloen/alpaca-lora
- https://github.com/huggingface/peft
- 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
- Linux or WSL
- Modern NVIDIA GPU with >= 16 GB of VRAM (but it might be possible to run with less for smaller sample lengths)
I recommend using a virtual environment to install the required packages. Conda preferred.
conda create -n simple-llm-finetuner python=3.10
conda activate simple-llm-finetuner
conda install -y cuda -c nvidia/label/cuda-11.7.0
conda install -y pytorch=2 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-llm-finetuner.git
cd simple-llm-finetuner
pip install -r requirements.txt
Launch it
python app.py
Open https://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 new LoRA adapter name in the "New PEFT Adapter 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/
directory.
After training is done, navigate to "Inference" tab, select your LoRA, and play with it.
Have fun!
https://www.youtube.com/watch?v=yM1wanDkNz8
MIT License