Orchestrate SSH-based ML experiment batches with stateful queueing and OOM-aware retries.
npx skills add https://github.com/jkfee/Auto-Research --skill experiment-queue
Please help me install this Agent Skill. Skill: experiment-queue Source: https://github.com/jkfee/Auto-Research/tree/main/skills/experiment-queue Command: npx skills add https://github.com/jkfee/Auto-Research --skill experiment-queue
SSH-based orchestration of large batches of ML experiments enables scalable, reproducible runs on remote GPU clusters by coordinating jobs, tracking state, and handling failures automatically.
Describe your experiment grid and run /experiment-queue to launch a scheduled, wave-based batch across multiple GPUs.
High-intent search queries and answers about installing and using this skill.
You can orchestrate large ML experiment grids by running scheduled, wave-based batches across multiple remote GPUs. The system coordinates SSH-based jobs, tracks persistent queue state, and handles failures automatically.
To recover ML experiments from CUDA out of memory errors, the system uses OOM-aware retries. It automatically recovers from memory errors during wave transitions while preserving prior experiment progress.
Wave-based scheduling for machine learning jobs is a method to maximize remote GPU utilization without memory contention. It processes batches of experiments in sequential waves to ensure stable resource allocation.
You run multi-seed and multi-config grid searches by parsing concise manifest specifications. The orchestrator generates and executes large experiment grids from these manifests with persistent state tracking.
Yes, you can restart interrupted SSH batch jobs without losing progress. The orchestration system enforces persistent queue state and robust restartability, allowing recovery from detached or stale screen sessions.
You should use a dedicated experiment queue instead of a single run command when handling large grid expansions, teacher-student phase workflows, or multi-seed configurations that exceed standard execution capacity.
wangdepinConfigure SLURM, RunPod, or local GPU clusters in .raca/clusters.yaml.
lightrain-aOrchestrate SSH-based ML experiment batches with wave and dependency enforcement.
auerbachbGenerate a clipboard-ready PM handoff prompt from repository config, GitHub state, and memory.
duypham2801Orchestrate SSH-hosted batch ML experiments with wave scheduling and OOM-aware retries.
