ML Training Recipes
Gives the agent proven PyTorch training patterns, optimizer and LR guidance, and debugging steps
- Repository
- Orchestra-Research/AI-Research-SKILLs
- Stars
- 14,195
- Licence
- MIT
- Files
- 7 files
What it does
This is a set of expert notes that teaches an AI agent how to train and fine-tune neural networks with PyTorch. Once installed, when you ask for help with training, the agent follows established practices for choosing a model for your data size, picking optimizers such as AdamW or Muon, setting learning-rate schedules, using mixed precision, and diagnosing problems like loss spikes or out-of-memory errors. Reference files cover domains from LLMs and vision to medical imaging and protein or genomics models.
Great for
- ML engineers fine-tuning or training models in PyTorch who want a checklist of sound defaults
- Researchers debugging diverging loss, memory errors or slow GPU throughput
- Students learning practical training recipes beyond textbook theory
Skip it if
- Non-programmers or people who only need to use a finished AI tool, since the content assumes PyTorch and GPU experience
- Production-critical decisions without expert review: the recipes are general patterns, and the agent may apply hyperparameters that do not suit your data or hardware
Try asking
- “My fine-tuned ViT loss spikes after a few hundred steps, what should I check?”
- “Which architecture should I use for 5,000 labelled medical images?”
- “Set up an AdamW plus Muon optimizer config for a small GPT-style model in PyTorch.”
- “I'm getting CUDA out-of-memory errors during training, how can I reduce memory use?”
Install in 3 steps
- 1
Download the skill
The ZIP contains the 7 original files, the MIT licence and an attribution note.
ml-training-recipes.zip - 2
Add it to your agent
Claude Code (all projects)
mkdir -p ~/.claude/skills && unzip ml-training-recipes.zip -d ~/.claude/skills/Claude Code / Agent SDK (this project only)
mkdir -p .claude/skills && unzip ml-training-recipes.zip -d .claude/skills/Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.
- 3
Ask for the task
No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.
Get better results
- Give the agent your dataset size, data type, and hardware so it can pick the right row in the recipe tables.
- Treat the hyperparameter values (such as LR 0.04 and weight decay 0.2) as starting points and validate them on your own data.
- Ask the agent to explain why it recommends a change so you can check it against the referenced papers and codebases.
What's inside
- references/architecture.md
- references/biomedical.md
- references/domain-specific.md
- references/experiment-loop.md
- references/optimizers.md
- references/scaling-and-selection.md
- SKILL.md
Redistributed unmodified from Orchestra-Research/AI-Research-SKILLs under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.