Llama Classification and Generation Templates

This repository provides templates of Llama models for a variety of natural language processing (NLP) tasks, including:

• Binary classification: Both zero-shot and fine-tuned classification tasks.
• Text generation: Leveraging Llama's local inference capabilities to generate text based on prompts.
• Model training with LoRA: Fine-tune Llama models efficiently using Low-Rank Adaptation (LoRA), which allows for model adaptation without retraining the entire model.

Key Features:

• Pre-built scripts for training Llama models on classification tasks and text generation.
• Fine-tuning capability for classification tasks using LoRA, enabling faster and more efficient adaptation.
• Support for local inference with Llama, making it easy to run the model on your own machine without relying on APIs or cloud-based services.

Setup & Requirements:

• Kaggle GPUs (preferably T4) are recommended for model training due to hardware requirements.
• Ensure you have access to the Llama models on Hugging Face to download them locally.

How It Works:

• The Llama model is downloaded and run locally on your machine.
• Training and inference scripts are provided for both classification and text generation tasks.
• LoRA is used for efficient model fine-tuning, allowing for the adaptation of the model without requiring full retraining.