Remote GPU Trainer

AutoDL/vast.ai/RunPod/Lambda/Paperspace远程GPU启动、调试、监控、续训、并行、停止及卡住/OOM/NaN/全零指标排查；可验证状态/selection-best checkpoint/拉回结果。释放实例销毁机器及数据盘，不等同关机。长期镜像/恢复→mirror-research-artifacts。

42 stars 6 forksResearch & Science

Install

skills CLI (Claude Code, Cursor, Codex and more)

npx skills add Hanyuyuan6/remote-gpu-trainer --skill remote-gpu-trainer

Manual install

git clone --depth 1 https://github.com/Hanyuyuan6/remote-gpu-trainer.git /tmp/remote-gpu-trainer
mkdir -p ~/.claude/skills
cp -r /tmp/remote-gpu-trainer/. ~/.claude/skills/remote-gpu-trainer

Personal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.

From the SKILL.md

# Remote GPU Trainer

## Mission

Move one experiment through **RUN → VERIFY → DELIVER** with the least control work that safely advances the
user's outcome. This is the compute/control layer and owns one compute attempt. When the task spans several
runs, nodes, independent side-effect lanes, or paper-wide closeout, route scheduling and state projection to
`research-artifact-hygiene` (closeout ledger); do not duplicate a second controller here. Companion skills named
anywhere in this skill (`research-artifact-hygiene`, `mirror-research-artifacts`, `superpowers:*`,
`huggingface-skills:*`) are optional separate installs: without one, do the equivalent step with your own tools.

## Start with the outcome, not a package

Before tools, write a compact internal decision:

1. desired result and current verified state;
2. exact next executable action;
3. consequence level below;
4. stop condition and where the result will live.

Do not preload incident catalogues, every platform profile, or a whole project SSoT. Pick one profile and one
phase reference. Keep large evidence out of model context: verify identity/schema, parse task-relevant fields,
and retain paths/hashes as an evidence index.

## Consequence-based assurance

Classify the next consequence, not the noun used in a handoff:

| Level | Examples | Required assurance |
|---|---|---|

Read the full skill on GitHub

Details

Skill name: remote-gpu-trainer
Repository: Hanyuyuan6/remote-gpu-trainer
License: MIT
Repo updated: 28 October 2026

FAQ

What does the Remote GPU Trainer skill do?

How do I install the Remote GPU Trainer skill?

skills CLI (Claude Code, Cursor, Codex and more): npx skills add Hanyuyuan6/remote-gpu-trainer --skill remote-gpu-trainer. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.

Is Remote GPU Trainer popular?

It lives in Hanyuyuan6/remote-gpu-trainer, which has 42 GitHub stars and 6 forks.

Related skills

Suoha Serenity Skill

suoha888/suoha-serenity-skill

Research stocks and supply-chain chokepoints using a framework distilled from Serenity's public posts. Use for evidence-first company health checks across A-shares, Hong Kong and US equities, live data packs, company-name resolution and falsifiable, source-cited research reports. Distinguish an industry bottleneck from a company's value capture and market expectations. Reports do not represent Serenity's own views; research only, no trade execution or promised returns.

Research & Science

Paper Workflow

brycewang-stanford/Paper-WorkFlow

将社科实证研究从 idea、proposal、数据或已有结果推进到可复现分析和完整论文 DOCX。支持研究设计、数据审计、估计、表图、写作与修订，以及按需准备投稿包；按证据完成阶段，不承诺显著结果或录用。触发：/paper-workflow、从想法到论文、从数据到 docx、推进已有实证稿件。

Research & Science

Bettersearch

ry0narushima/bettersearch

Web research with DeepSeek's Search & Fetch APIs. Use whenever the user asks to research something, look up or look into a topic, search the web, find sources, check current facts, numbers or prices, compare options with up-to-date info, fact check a claim, or do deep research / write a research report. Quick mode answers in chat with numbered citations; advanced (deep) mode runs a planner, researchers, fact-checker, reviewer and writer and builds a self-contained HTML report.

Research & Science

Synthetic Market Research

BayramAnnakov/synthetic-market-research

Fast, cheap market research using LLM-generated synthetic survey responses with Semantic Similarity Rating (SSR). Runs purchase intent, concept tests, and pricing research in minutes instead of weeks, at $0 per respondent. Based on PyMC Labs' validated methodology (90% correlation with real humans across 57 surveys). Triggers: synthetic research, market research, survey, purchase intent, concept test, consumer research, SSR, synthetic survey, product validation, pricing research, Likert scale.

Research & Science

Paper Sprint Review

RichradsY/PaperSprint

PaperSprint is a Scrum-inspired academic paper agent workflow for conference or journal drafts. Use when Codex needs to clarify paper direction, estimate likely sprint count, set sprint goals and focus areas, assess venue fit, organize reviewer or editor lenses, convert critique into a prioritized revision backlog, drive amendment increments, or keep a process log across repeated review loops for `.docx`, `.tex`, `.md`, PDF, thesis, and response-letter materials.

Research & Science

Labyrinth Exploration

nasqret/labyrinth-exploration

Map and advance open-ended research programmes in mathematics, theoretical science and algorithms. Use for pushing bounds, attacking open conjectures, mapping what is still unknown, or independently refereeing and integrating research from agents or external models. Maintain proof tiers, provenance, frontier data and a dashboard. Not for a single self-contained proof, a one-off literature question or ordinary code review.

Research & Science

Browse the most popular Claude skills, all research & science skills, or write your own with the skills generator.