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