autoresearch

Autonomous ML experiments on a remote VPS require an orchestration loop that executes iterative training, evaluates results, and patches code via SSH. This skill coordinates these steps using a user-supplied session file to drive autonomous runs and record metrics.

An autonomous ML experiment loop iteratively proposes code changes, executes training on a remote VPS, and parses results to evaluate hypotheses. It automatically commits successful changes or rolls back failed ones while maintaining a reproducible experiment history.

To start an autonomous experiment session, place a session.md file at the default path to define constraints and guidelines. The agent reads this session file to orchestrate training, fetch and edit code, and run evaluations on the VPS automatically.

Yes, SSH access is required to run iterative model training on a VPS using this skill. It relies on SSH-driven orchestration to fetch and patch remote training scripts, execute the training process, and handle automatic code commits or rollbacks.

Automated ML training handles failed experiments by parsing training results and automatically rolling back code changes. It enforces session-defined constraints and guardrails to maintain a reproducible experiment history and preserve stable training code.