bitsandbytes-foundation/bitsandbytes

bitsandbytes is a PyTorch library providing 8-bit and 4-bit quantization and optimizers for reducing memory use in large language model training and inference

View on index

Related repos

• Code and tutorials for building large language models from scratch in PyTorch
• minimal PyTorch code for training your own small language model from scratch
• DeepSeek-V3, a Mixture-of-Experts language model with 671B parameters and 37B active per token, focused on math and coding tasks
• PyTorch deep learning course files and code notebooks
• reinforcement learning library for training large language models, supports many RL algorithms, integrates with LLM tools, scales to models with up to 671B parameters
• PyTorch toolkit for speech recognition, speaker ID, text-to-speech, and conversational AI tasks
• RAG with memory - processes up to 1M tokens and generates contextual clues from global understanding
• run large language models on any device
• RNN meets transformer for efficient large language models
• tensor library for machine learning with support for hardware acceleration and model inference