#6060th of 105 in ML frameworks and libraries
bitsandbytes
8-bit and 4-bit quantization and 8-bit optimizers for PyTorch models
- Stars
- 8.5k
- License
- MIT
- Last commit
- Sep 2026
- Last release
- Aug 2026
- Language
- Python
8,030 on 1 Apr 20268,529 on 21 Oct 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
bitsandbytes is a PyTorch library that provides LLM.int8() 8-bit inference, QLoRA 4-bit quantization and block-wise 8-bit optimizers. It exposes Linear8bitLt and Linear4bit layers plus a bitsandbytes.optim module, and is documented as a quantization backend for Hugging Face Transformers, Diffusers and PEFT. Backends cover CPU, NVIDIA, AMD, Intel GPU and Gaudi, and Apple Metal.
Who it is for: ML engineers fitting or fine-tuning large models in less GPU memory
Strengths
- Drop-in Linear8bitLt and Linear4bit layers plus 8-bit optimizers in bitsandbytes.optim
- Documented integrations with Hugging Face Transformers, Diffusers and PEFT
- Wide hardware matrix: x86-64 and aarch64 CPU, NVIDIA, AMD, Intel XPU, Windows, macOS
- Backed by published papers: LLM.int8(), QLoRA and 8-bit optimizers
Weaknesses
- Library only, no model server, UI or CLI; requires writing PyTorch code
- Gaudi has partial 4-bit support and no 8-bit optimizers
- Metal 8-bit optimizers only planned; some CPU and Metal paths lack optimizations
- Requires Python 3.10+ and PyTorch 2.4+
What it needs
- GPU optional
- Needs PyTorch 2.4+, Python 3.10+
- Models: PyTorch models, Hugging Face Transformers, Diffusers, PEFT
Also in ML frameworks and libraries
See all 105| Rank | Project | Score |
|---|---|---|
| 1 | transformersPython library defining open models for text, vision, audio and multimodal tasks | 94 out of 100 |
| 2 | pytorchPython tensor library with GPU support and tape-based autograd for deep learning | 94 out of 100 |
| 3 | ultralyticsPython library and CLI for training and running YOLO vision models | 93 out of 100 |
| #4 | pytorch-image-modelsPyTorch library of image models, pretrained weights and reference training scripts | 92 out of 100 |
| #5 | supervisionPython toolkit for loading, annotating and counting computer vision detections | 91 out of 100 |