Maple-Preview
Maple-Preview is a 20B-parameter mixture-of-experts language model using ternary weights, built for efficient on-device reasoning. It supports a 131,072-token context window and is designed to run fast on consumer hardware such as Apple Silicon Macs and iPhones, reportedly reaching over 200 tokens per second on an M4 Mac mini. The model focuses on reasoning tasks, including competition-level math and coding, while remaining lightweight enough for local deployment.
- Type
- Text
- Weights
- Open
- Downloads
- 2.2k
- Likes
- 413
- Pipeline
- text generation
- License
- mit
Model hub
deepgrove/maple-preview
- Repo
- deepgrove/maple-preview
- Library
- transformers
- Access
- Public
Technical tags and languages
More from DeepGrove
3
About DeepGrove
1
Independent AI research lab building efficient, ternary-weight language models designed to run frontier intelligence on any device (edge computing).