elixir-nx/bumblebee

Bumblebee is a machine learning library for the BEAM that provides a unified interface for loading, configuring, and running pre-trained models for text, vision, and audio tasks. It supports a wide range of architectures, including transformers for NLP, diffusion models for image generation, and speech-to-text pipelines, all accessible through high-level serving APIs. The system handles model conversion from PyTorch and Hugging Face formats, manages tokenization and preprocessing, and integrates with the Phoenix framework for web-based inference.

This period focused on expanding the library's capabilities to support a wide range of multimodal and generative tasks, including text, vision, audio, and image generation. The work introduced new model architectures for diffusion models (Stable Diffusion, ControlNet) and various transformer-based models, alongside comprehensive test coverage and utility modules to support these features.

28 changes

This period focused on integrating the Whisper speech-to-text model into the Bumblebee framework, including its serving pipeline and associated test coverage. It also introduced extensible logits processing for text generation and added support for loading PyTorch model states directly into the system.

7 changes

elixir-nx/bumblebee was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point the surveyor at a repository you know and see whether you agree with it.