Machine learning

Formats for trained models. Each stores the model’s weights as large arrays of numbers called tensors; some also describe the network that uses them, so another program can run the model.

Formats

.GGUF

GGUF model file

A language model for llama.cpp: metadata and quantised weights.

.ONNX

Open Neural Network Exchange

A neural network’s graph and weights in one Protocol Buffers message.

.SAFETENSORS

Safetensors

Model weights after a JSON header, safe to load with no code.