A 1.6B decoder-only LM whose training can be independently verified, with midtrained and SFT variants. Record: https://open1b.gensyn.ai
AI & ML interests
We network together the core resource for machine intelligence to flourish alongside human intelligence https://www.gensyn.ai/research
Recent Activity
Papers
The Role of Feedback Alignment in Self-Distillation
IR3DE: A Linear Router for Large Language Models
-
Training-Free Dynamic Upcycling of Expert Language Models
Paper • 2603.29765 • Published • 10 -
Sharing is Caring: Efficient LM Post-Training with Collective RL Experience Sharing
Paper • 2509.08721 • Published • 664 -
All is Not Lost: LLM Recovery without Checkpoints
Paper • 2506.15461 • Published • 41 -
NoLoCo: No-all-reduce Low Communication Training Method for Large Models
Paper • 2506.10911 • Published • 9
A 1.6B decoder-only LM whose training can be independently verified, with midtrained and SFT variants. Record: https://open1b.gensyn.ai
-
Training-Free Dynamic Upcycling of Expert Language Models
Paper • 2603.29765 • Published • 10 -
Sharing is Caring: Efficient LM Post-Training with Collective RL Experience Sharing
Paper • 2509.08721 • Published • 664 -
All is Not Lost: LLM Recovery without Checkpoints
Paper • 2506.15461 • Published • 41 -
NoLoCo: No-all-reduce Low Communication Training Method for Large Models
Paper • 2506.10911 • Published • 9
models 10
Gensyn/open-1b-sft-mlx-bf16
Text Generation • 2B • Updated • 201
Gensyn/open-1b-sft
Text Generation • 2B • Updated • 1.15k • 16
Gensyn/open-1b-midtrained-93B
Text Generation • 2B • Updated • 635 • 4
Gensyn/open-1b-base
Text Generation • 2B • Updated • 1.94k • 7
Gensyn/testing
112k • Updated • 8
Gensyn/Qwen2.5-7B-Instruct
Text Generation • 8B • Updated • 23 • • 7
Gensyn/Qwen2.5-72B-Instruct-bnb-4bit
Text Generation • 75B • Updated • 7
Gensyn/Qwen2.5-32B-Instruct-bnb-4bit
Text Generation • 34B • Updated • 7
Gensyn/Qwen2.5-1.5B-Instruct
Text Generation • 2B • Updated • 36 • • 17
Gensyn/Qwen2.5-0.5B-Instruct
Text Generation • 0.5B • Updated • 131 • • 33
datasets 0
None public yet