NLP Pretraining Best Practices
Guides the agent to write and review language model training code using standard fine-tuning recipes.
- Repository
- aiming-lab/AutoResearchClaw
- Stars
- 14,802
- Licence
- MIT
- Files
- 1 file
What it does
This is a set of written guidelines that an AI agent reads when it is asked to write or review code for training or fine-tuning language models. Once installed, the agent follows standard recommended settings, such as learning rates, batch sizes, warmup schedules, and parameter-efficient methods like LoRA, and evaluates results with common metrics and multiple random seeds. It contains no executable code of its own; it only shapes the agent's advice and code.
Great for
- Researchers or engineers writing fine-tuning scripts for BERT- or GPT-style models
- Reviewers checking whether an existing NLP training setup follows common defaults
- Students learning the typical hyperparameters and evaluation practices for language models
Skip it if
- Production decisions that need validated results: the recipe gives generic ranges, not tuned values for a specific dataset, and the agent may still produce wrong code
- Non-technical users wanting a finished model: the skill assumes familiarity with training pipelines and does not install or run anything
Try asking
- “Write a fine-tuning script for a BERT classifier on my support tickets dataset”
- “Review my training config and tell me if the learning rate and warmup look reasonable”
- “Set up LoRA fine-tuning for a small GPT model with sensible rank and alpha values”
- “What metrics should I report for an imbalanced text classification task?”
Install in 3 steps
- 1
Download the skill
The ZIP contains the 1 original files, the MIT licence and an attribution note.
nlp-pretraining.zip - 2
Add it to your agent
Claude Code (all projects)
mkdir -p ~/.claude/skills && unzip nlp-pretraining.zip -d ~/.claude/skills/Claude Code / Agent SDK (this project only)
mkdir -p .claude/skills && unzip nlp-pretraining.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, task type and hardware so it can adjust the generic ranges to your situation.
- Check any generated code before running it, especially checkpoint loading from the HuggingFace hub and the seed and evaluation logic.
- Ask for the reasoning behind each hyperparameter so you can validate it against your own experiments.
What's inside
- SKILL.md
Redistributed unmodified from aiming-lab/AutoResearchClaw 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.