pytorch-fsdp

Expert guidance for Fully Sharded Data Parallel training with PyTorch FSDP - parameter sharding, mixed precision, CPU offloading, FSDP2

Quality

48%

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

Fix and improve this skill with Tessl

tessl review fix ./backend/cli/skills/ml-training/pytorch-fsdp/SKILL.md

Quality

Content

15%

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

The body is an auto-generated documentation dump rather than curated skill guidance: it is extremely verbose, lacks executable code and any real workflow, and inlines content that belongs in well-organized reference files.

Suggestions

Replace the inlined 130K-character doc dump with a concise overview and move detailed material into topical reference files (e.g. references/sharding.md, references/mixed-precision.md), keeping SKILL.md lean.

Add concrete, complete, copy-paste-ready code examples for the core FSDP setup flow (wrap model, init process group, training loop) instead of bare-token code blocks.

Introduce a clear multi-step workflow with validation checkpoints (e.g. verify process group, check sharding, validate gradients) for common FSDP setup and debugging tasks.

Dimension	Reasoning	Score
Conciseness	The body dumps full official documentation pages inline (e.g. a single 130,541-character line for Pattern 2) including "Created On / Last Updated On" timestamps and explanations of concepts like torch.distributed backends that Claude already knows — a verbose, padded wall of text.	1 / 3
Actionability	There are some concrete REPL snippets (e.g. `>>> mesh_2d = init_device_mesh("cuda", mesh_shape=(2, 8), mesh_dim_names=("dp", "tp"))`), but many code blocks contain only a bare token (`Join`, `NCCL`, `True`) rather than complete executable examples, so guidance is incomplete.	2 / 3
Workflow Clarity	There is no multi-step process sequence and no validation checkpoints; the content is a flat reference dump of patterns rather than a guided workflow, matching the score-1 anchor of unclear steps and no validation.	1 / 3
Progressive Disclosure	The SKILL.md is a monolithic wall of pasted documentation (the body should be an overview), and the only referenced bundle file is a generic `other.md` with no topical breakdown or clear navigation — the classic monolithic/nested-reference anti-pattern.	1 / 3
	Total	5 / 12 Passed

Description

82%

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 specific and distinctive, naming concrete FSDP capabilities with natural trigger terms, but it omits an explicit "Use when..." clause, which caps completeness at 2.

Suggestions

Append an explicit trigger clause, e.g. "Use when setting up FSDP/FSDP2 training, configuring sharding or mixed precision, or debugging distributed training issues."

Confirm it is not the level below for specificity by ensuring each named capability (sharding, offloading) maps to a real, distinct action rather than a buzzword.

Dimension	Reasoning	Score
Specificity	Quotes "parameter sharding, mixed precision, CPU offloading, FSDP2" — it lists multiple concrete, specific capabilities rather than vague language, matching the score-3 anchor.	3 / 3
Completeness	It clearly states what the skill does but has no "Use when..." or equivalent explicit trigger clause; per the rubric, a missing explicit trigger caps completeness at 2, so it cannot reach 3.	2 / 3
Trigger Term Quality	Terms like "Fully Sharded Data Parallel", "PyTorch FSDP", "mixed precision", "CPU offloading", and "FSDP2" are exactly what a user would naturally say when needing this skill, giving good coverage of natural terms.	3 / 3
Distinctiveness Conflict Risk	The niche (FSDP/FSDP2 distributed training) is clearly defined with distinct triggers, making it unlikely to fire for unrelated skills; not the level below because it is not merely "somewhat specific."	3 / 3
	Total	11 / 12 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: synthetic-sciences/openscience
Path: backend/cli/skills/ml-training/pytorch-fsdp/SKILL.md
Commit: dc00e07

Reviewed: about 17 hours 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.