x.comthreadUSA SuperIntelligence | USASI (@USASIHQ) on X
Sevren released Heimr 570M, an experimental hybrid Mixture-of-Experts model designed for on-device use on phones and laptops, with about 570M active parameters out of 2.7B total. It is trained for a 64K context window to support long-context local decoding, targeting edge-AI deployment rather than cloud serving.
TL;DR Sevren's Heimr 570M is a small hybrid-MoE model (570M active / 2.7B total params, 64K context) built for local inference on phones and laptops.
Organisations named here
- Sevren
Technologies named here
- Heimr 570M
- Mixture of Experts
Techniques named here
- Hybrid MoE
Related in models
