Meta
Meta's AI research division is best known for the Llama 2 family of LLMs but has released nearly 2,000 open source models across a variety of use cases.
Meta models
About Meta
Meta’s AI research divisions create a wide range of models, with almost 2,000 models available. These models include the Llama 2 family of large language models, the AudioGen family of audio creation models, and a variety of useful utility models for tasks from image segmentation to text encoding.
What license does Meta release models under?
Meta’s most famous model family, Llama 2, has a custom license that allows for commercial use in some circumstances. Because of this license, you must register with Meta before downloading model weights from Hugging Face or deploying Llama models to Baseten.
Many of their other models, including the AudioGen model family, are licensed only for non-commercial use under cc-by-nc-4.0.
However, some models, such as their Segment Anything Model (SAM) family, are licensed for commercial use with licenses like Apache 2.0. Always check the license of the model you’re planning to use before deploying it.
What are Meta models used for?
Meta’s Llama family of LLMs includes base models, chat-tuned models, and CodeLlama models specifically designed for code completion and code chat. Llama models can be used as a replacement for proprietary LLMs like GPT-3.5.
Meta’s vast library of utility models are useful for a wide range of classification, segmentation, feature extraction, and object detection tasks.
Can I fine tune Meta models?
Meta’s Llama family of LLMs are foundation models with base variants that serve as a great starting point for fine tuning. Llama 2 is one of the most popular starting points for fine tuning LLMs for evaluation benchmarks, with model families like WizardLM using Llama 2 as a starting point.
