moe 4
- FreeToken: 284B on a Gaming Desktop, 753B on One Workstation GPU - and the Ratio That Makes It Work
- LocalCowork Architecture: Sub-Second Tool Routing with On-Device MoE Models
- Qwen3.5-Medium Architecture: MoE Routing, 1M Context, and Sonnet-Class Local Inference
- Source Audit: KTransformers and Heterogeneous CPU-GPU MoE Inference