ByteShape is an AI company specializing in the quantization of large language models, the technique of compressing model weights to fewer bits to reduce memory use. It publishes GGUF files of open models compressed to low bit widths with its own ShapeLearn technology on Hugging Face.
The models ByteShape releases are intended for local inference servers such as llama.cpp, with the aim of running large models on a single consumer-grade GPU.