Edge0-35B
Edge0-35B is a 35 billion parameter sparse mixture of experts language model built on Qwen3.5-MoE 35B-A3B, designed to run fully on-device on consumer hardware. It uses 4-bit quantization, SSD expert offload, and a trained prerouter to stream expert weights from storage, keeping peak memory near 3GB while sustaining interactive generation speeds on Apple Silicon.
- Type
- Text
- Weights
- Closed
- Downloads
- 37k
- Likes
- 3.3k
- Pipeline
- text generation
- License
- apache-2.0
Model hub
Edge0/Edge0-35B-A3B-preview
- Repo
- Edge0/Edge0-35B-A3B-preview
- Library
- mlx
- Access
- Public
Technical tags and languages
Papers
1
More from Edge0
6
About Edge0
1
Edge0 builds efficient foundation models and an open-source streaming mixture-of-experts inference framework for private, offline AI on consumer hardware.