ml-training-recipes

Battle-tested PyTorch training recipes for all domains — LLMs, vision, diffusion, medical imaging, protein/drug discovery, spatial omics, genomics. Covers training loops, optimizer selection (AdamW, Muon), LR scheduling, mixed precision, debugging, and systematic experimentation. Use when training or fine-tuning neural networks, debugging loss spikes or OOM, choosing architectures, or optimizing GPU throughput.

Quality

96%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Quality

Content

93%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A well-engineered recipe reference: lean and actionable with executable code, well-sequenced checklists carrying validation cues, and clean one-level-deep progressive disclosure into real bundle files. The only soft spot is workflow clarity, since guidance is distributed across many parallel sub-processes rather than one coherent validated workflow.

Dimension	Reasoning	Score
Conciseness	Dense, high-signal content that assumes Claude's competence — no padding explaining what Transformers or PyTorch are; even notes earn their place (e.g., 'eps=1e-10 ... prevents stale updates in bf16').	5 / 5
Actionability	Copy-paste-ready executable code throughout (training loop, cosine/WSD schedulers, meta-device init, MFU calc) plus specific value tables (optimizer LR/WD per parameter type, Chinchilla token budgets).	5 / 5
Workflow Clarity	Sequenced checklists and ordered troubleshooting flows (OOM solutions 'in order', debugging steps, hyperparameter priority) with validation signals (fast-fail on loss>100, keep/discard/revert loop), though no single end-to-end workflow ties every checkpoint together.	4 / 5
Progressive Disclosure	A reference-file table up top with one-line descriptions, inline pointers at the right moments ('For biomedical domains, see references/biomedical.md'), and verified one-level-deep references — core cross-domain recipes inline, specialized detail split out.	5 / 5
	Total	19 / 20 Passed

Description

100%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

The description is exemplary: it states concrete capabilities with named techniques, provides a rich explicit 'Use when' trigger clause, and carves out a distinct niche with low conflict risk, all in third person without padding.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions with named techniques — 'training loops, optimizer selection (AdamW, Muon), LR scheduling, mixed precision, debugging, and systematic experimentation' — giving comprehensive coverage rather than generic verbs.	5 / 5
Completeness	Explicitly answers both: 'what' (recipes covering training loops, optimizer selection, scheduling, precision, debugging) and 'when' via a concrete 'Use when ...' clause enumerating trigger scenarios.	5 / 5
Trigger Term Quality	Natural user phrases abound — 'training or fine-tuning neural networks', 'debugging loss spikes or OOM', 'choosing architectures', 'optimizing GPU throughput' — with synonyms (training/fine-tuning) and concrete file/format-adjacent terms users actually say.	5 / 5
Distinctiveness Conflict Risk	Occupies a clear niche (PyTorch training recipes across ML domains) with distinctive triggers like 'loss spikes', 'OOM', 'GPU throughput', 'Muon/AdamW', making conflict with unrelated skills minimal.	5 / 5
	Total	20 / 20 Passed

Validation

93%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 15 / 16 Passed

Validation for skill structure

Criteria	Description	Result
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	15 / 16 Passed

Repository: OpenRaiser/NanoResearch
Path: skills/vendor-ai-research/ml-training-recipes/SKILL.md
Commit: 9d3b440

Reviewed: 11 days ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.