Llama 3.2 3 B Instruct
unslothIntroduction
The Llama 3.2-3B-Instruct is a multilingual, instruction-tuned large language model developed by Meta. It is part of the Llama 3.2 collection, optimized for generative tasks such as dialogue and summarization. It is designed to outperform many open and closed source chat models in industry-standard benchmarks.
Architecture
Llama 3.2 is an auto-regressive language model utilizing an optimized transformer architecture. It includes various multilingual capabilities, supporting languages such as English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai. The model is trained using supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to ensure alignment with human preferences for helpfulness and safety.
Training
The Llama 3.2 models have been pretrained and instruction-tuned on a diverse dataset. They incorporate techniques like Grouped-Query Attention (GQA) to enhance inference scalability. The model supports fine-tuning beyond its supported languages, adhering to the Llama 3.2 Community License and Acceptable Use Policy.
Guide: Running Locally
- Setup Environment: Ensure you have Python and necessary libraries such as
transformers
installed. - Download Model: Access the model via Hugging Face or the provided Google Colab notebooks.
- Run Notebook: For a quick start, use the free Google Colab Tesla T4 notebook for Llama 3.2 (3B) here.
- Fine-tuning: Customize the model with your dataset by following the instructions in the notebook.
- GPU Recommendation: Utilize cloud GPUs, such as Google Colab's Tesla T4, for efficient processing.
License
The Llama 3.2 model is distributed under the Llama 3.2 Community License, a custom commercial license agreement. This license governs the use and distribution of the model. More details can be found in the license document.