Meta has released artificial intelligence models in an open format for research purposes only, including:
Chameleon, a mixed-modal AI model that can process both images and text, allowing users to input commands to generate images along with sample images. The model can also generate images in response to text inputs. This type of model may pave the way for new usage formats, such as creating images with continuous content related to the original input.
Multi-Token Prediction, an LLM model architecture that can predict multiple future words at a time, accelerating the work process compared to traditional LLM models that generate responses word by word. The current release of this model is intended for code completion assistance only.
JASCO, a model that generates music from various inputs, including regular text, song chords, and rhythms. This model provides users with more control over the creative process.
These models are available for download for experimentation purposes only, to replicate the research results published by Meta’s team. While some models, like the high-speed code completion model, may have practical uses, they are not available for commercial use.
Source: Meta Newsroom
TLDR: Meta has released AI models for research purposes, including Chameleon for mixed-modal processing, Multi-Token Prediction for fast word predictions, and JASCO for music generation. Models are not for commercial use.
Leave a Comment