Adaption opens AutoScientist’s automated training loop as an API and SDK
sarahookr · x · 2026-07-28
Adaption released the AutoScientist API and SDK, bringing its automated model-training loop into code.
The system takes a dataset and a target objective, then iterates across four stages:
- search training recipes,
- optimize data,
- train,
- evaluate.
According to the post, AutoScientist can outperform human-configured training by up to 35% relative improvement and stops when it reaches the target or exhausts the budget. The new API lets developers trigger and monitor the loop programmatically from notebooks or pipelines, with the first line of the workflow starting from data upload and dataset adaptation.
Related event: Adaption Launches AutoScientist API(2 posts)→
More from Infra
- Streaming 1.6TB of K3 weights on an M5 Max 128GB is still slow — antirez · 2026-07-29
- Intent Lab says it sped up GLM 5.2 inference 6.3x on 2× Grace Blackwell — jiayq · 2026-07-29
- Recursive Superintelligence signs a $410 million compute deal with AWS — ChengleiSi · 2026-07-29
- Intel brings its 14A process forward to 2027 risk production and 2028 volume ramp — Beth_Kindig · 2026-07-29
- Owning the datacenter, not the model, is the new AI dream — AravSrinivas · 2026-07-29
- NVIDIA points Jetson users to a guide for running GenAI locally on-device — NVIDIAAI · 2026-07-28