Llama
Website: https://Llama.com
Options: Chatbot, API, Local/Offline
Price: Free (downloadable)
The Llama family of LLM models are made by Meta. They offer open-source text and image LLMs that you can download and run on your own computer. These free downloadable models have played a huge role in the growth of the local LLM/indie hacker community.
Their flagship open-source model has 405B parameters and is used as the foundation for model training by other LLM companies. Other models include:
– 1B and 3B lightweight text LLM models
– 11B and 90B multi-modal LLM models
– ** NEW: 70B multi-modal LLM with similar performance to the Llama 3.1 flagship 405B model
While the main benefit is that anyone can download and use these models locally, there are also online LLM providers that host them for you — so you can easily access them online via a standard chatbot interface.
Reddit Discussions about LLaMa
11/29 – I’ve made an “ultimate” guide about building and using `llama.cpp`
11/25 – Speculative decoding just landed in llama.cpp’s server with 25% to 60% speed improvements
11/16 – Nvidia presents LLaMA-Mesh: Generating 3D Mesh with Llama 3.1 8B. Promises weights drop soon.
11/4 – I succeeded in running Llama 3.1 405B after buying a little more RAM
11/4 – Accidentally Built a Terminal Command Buddy with Llama 3.2 3B model
11/1 – Testing llama.cpp with Intel’s Xe2 iGPU (Core Ultra 7 258V w/ Arc Graphics 140V)