Skip to content

Latest commit

 

History

History
77 lines (49 loc) · 2.06 KB

README.md

File metadata and controls

77 lines (49 loc) · 2.06 KB
Raw

Llama Prompt Generator

Llama Prompt Generator

This repository contains a demo script for running Meta Llama on Linux using PyTorch.

Requirements

  • Python
  • PyTorch
  • Transformers
  • Hugging Face CLI
  • bitsandbytes
  • datasets
  • PEFT
  • numpy

Setup

  1. Install Required Packages

    pip install torch transformers huggingface_hub[cli] bitsandbytes datasets peft numpy

  2. Login to Hugging Face CLI

    huggingface-cli login

Usage

Finetune the Model

Run the finetune_meta_llama.py script to finetune the model on the stable-diffusion prompts dataset.

python finetune_meta_llama.py

Generate Text

Run the generate_text.py script to generate text based on the prompts and themes. 🚀🚀

You can now use these updated commands for additional info.

  • --themes: Include themes by specifying common stable diffusion themes.
    • Example: --themes scifi
  • --output_count: Control the number of output logs by specifying the desired number.
    • Example: --output_count 1200
  • --include_metadata: Include additional metadata in the generated text outputs.
  • --log_generation: Enable detailed logging of the generation process.
  • --use_model_prompts: Generate prompts directly using the model instead of loading from tokenized_examples.json --seed: Set a random seed for reproducibility. Example: --seed 51544
python generate_text.py --themes scifi fantasy horror art --output_count 1200 --log_generation --include_metadata

🚀🚀

python generate_text.py --themes scifi fantasy horror art --output_count 1200 --log_generation --include_metadata --use_model_prompts --seed 51544

Dataset

This project uses the MetaLlama Text Generation Prompts dataset.

It's a prompt extender/enhancer based on: Stable Diffusion Prompts