Domyn releases five open specialist teacher models for Edge 4B distillation
·2026.11.12 00:39
Key point
The release includes Math, STEM, Code, Instruction Following, and Agentic models trained with GRPO on verifiable rewards.
Details
- Domyn released bundle 2 of the Domyn Edge project, comprising five open specialist teacher models: Math, STEM, Code, Instruction Following, and Agentic.
- Domyn Edge is a 4B model being trained from scratch on 13T tokens on EuroHPC, specifically designed for agentic work at long context.
- The five teachers were created from the released SFT checkpoint using Multi-Teacher On-Policy Distillation (MOPD), with each model undergoing an SFT pass followed by GRPO with verifiable rewards.
- All five teachers, their data mixes, and training recipes are available in the Domyn Edge collection on Hugging Face, alongside bundle 1 which contains the base model, pre-training checkpoints, SFT student, and training code.
- Bundle 3 is scheduled to include the final distilled model and a technical report.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.