Open-Gen released: open-source PyTorch reconstruction of Generalist's embodied foundation model
KyeGomezB · x · 2026-08-20
Kye Gomez has released Open-Gen, an independent open-source PyTorch reconstruction of Generalist AI's GEN embodied foundation model family, built from public evidence. The README restates the core design: a decoder-only transformer over continuous time where sensing and acting are asynchronous token streams at different clock rates. The load-bearing trick is one line of masking — an action token at time t may only attend to sensor tokens at t−δ, δ being the inference latency sampled during training — turning latency into a modelled quantity, so no System 1/System 2 split or inference-time guidance is needed to hit 100 Hz from a 7B model. The author notes he is unaffiliated with Generalist AI, and that real datasets, fused attention, μP scaling, the 1B→7B crossover, physical prompting, and real-robot experiments remain to be validated.
Related event: Open-Gen: Open-Source Reimplementation of Generalist's GEN 1.5(3 posts)→
More from Embodied
- LLM-Controlled Robots Have <10% Chance of Reliability by 2028 — a_karvonen · 2026-08-20
- Using RealSense 3D camera for virtual green screen with depth control — chrismatthieu · 2026-08-20
- GEN-1.5 success driven by repetitive motion data and UMI collection — DrJimFan · 2026-08-20
- NVIDIA Tutorial: Post-Train Cosmos 3 Edge for On-Device Robot Control on Jetson Thor — MonaJalal_ · 2026-08-20
- Open-Gen Attempts to Replicate GEN 1.5 Embodied Model Architecture — KyeGomezB · 2026-08-20
- Three seconds to learn: Gen1.5's few-shot skill learning may hinge on one attention masking rule — KyeGomezB · 2026-08-20