hhuytho
- Karma
- 0
- Created
- ()
- Submissions
- 0
-
This repo has optimized runtimes for three VLAs on Jetson Thor: ABC-VLA, MolmoAct2, and pi-0.5. We used our agentic search to propose the custom fused kernels, and the runtime glue between engines. Changes are only…
- Show HN: Bonsai 1.7B ternary model at 442T/s on M4 Max (agents2agents.ai)
We took a recently released Bonsai 1.7B ternary model from PrismML (https://github.com/PrismML-Eng/Bonsai-demo) and ran our agentic evolution search on it for 6 hours to optimize the Metal kernels. The search was fully…