Woof 2B MLX 4-bit v1.1

Woof is a fine-tuned text model for browser tool use, packaged for local inference on Apple Silicon with MLX.

Model details

• Release: ConwayResearch/woof-2B-mlx-4bit-v1.1
• Parameters: 2,516,756,480 (approximately 2.52 billion).
• Precision: Affine 4-bit quantization with group size 64.
• Download size: approximately 1.43 GB.
• Input: text, including browser page observations and tool results.
• Output: reasoning, structured tool calls, and final responses.

This release contains merged model weights, configuration, tokenizer, and chat template. A separate training adapter is not required.

Usage

Download all model files and load the folder with an MLX-compatible tool that supports this model on Apple Silicon.

Other format

The same release is also available as Woof 2B MLX BF16 v1.1.