pytorch-fsdp2

Adds PyTorch FSDP2 (fully_shard) to training scripts with correct init, sharding, mixed precision/offload config, and distributed checkpointing. Use when models exceed single-GPU memory or when you need DTensor-based sharding with DeviceMesh.

Quality

78%

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 ./08-distributed-training/pytorch-fsdp2/SKILL.md

Quality

Content

57%

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

A well-structured, genuinely progressive skill body with a clear sequenced procedure and real one-level-deep references, but it leans on outline-level patterns over inline executable code and carries some reference repetition that costs tokens.

Suggestions

Consolidate the per-section "Reference:" lines into the existing final References section (or keep only one) to cut repeated tokens and lift conciseness.

Inline one small complete copy-paste example (e.g. a minimal init + bottom-up fully_shard + optimizer block) so actionability reaches fully executable rather than snippet-level.

Add an explicit validate-then-proceed checkpoint inside the checkpointing step (e.g. verify DCP metadata/coordinator before declaring a save successful) to embed feedback loops directly in the workflow.

Dimension	Reasoning	Score
Conciseness	The body is mostly lean and assumes Claude's competence (no basics about PyTorch/what FSDP is), but each section ends with a repeated "Reference:" line and a final References list duplicates many of them, adding tokens that could be tightened.	2 / 3
Actionability	Concrete API calls are given (init_process_group, fully_shard, MixedPrecisionPolicy, DCP helpers) but key patterns stay at the snippet/outline level (e.g. "iterate modules, if isinstance(m, TransformerBlock)") with complete runnable code deferred to bundle files rather than inline copy-paste examples.	2 / 3
Workflow Clarity	A clear 0-7 sequence plus Workflow A/B checklists and a debug feedback checklist exist, but validation checkpoints are not embedded inline in the main procedure for the risky checkpointing operations — they appear separate and somewhat implicit.	2 / 3
Progressive Disclosure	SKILL.md is an overview with well-signaled, one-level-deep references; all 12 referenced files exist, none reference further files, and detailed API/tutorial content is appropriately pushed to the references directory.	3 / 3
	Total	9 / 12 Passed

Description

100%

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

A concise, specific description that states concrete capabilities and an explicit use-when trigger in third-person voice, with domain-natural keywords and a clear distinct niche.

Dimension	Reasoning	Score
Specificity	Lists several concrete actions — "correct init, sharding, mixed precision/offload config, and distributed checkpointing" — rather than vague language, matching the multi-action anchor.	3 / 3
Completeness	Explicitly answers both what it does (adds FSDP2 with init/sharding/precision/checkpointing) and when to use it via an explicit "Use when..." clause, hitting the anchor for both what and when.	3 / 3
Trigger Term Quality	Includes natural terms a user would say for this need — "PyTorch FSDP2", "fully_shard", "single-GPU memory", "DTensor-based sharding", "DeviceMesh" — with good coverage of the domain's vocabulary.	3 / 3
Distinctiveness Conflict Risk	The FSDP2/fully_shard/DTensor/DeviceMesh niche is highly specific and unlikely to trigger for unrelated skills; uses third person ("Adds") as required.	3 / 3
	Total	12 / 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: Orchestra-Research/AI-Research-SKILLs
Path: 08-distributed-training/pytorch-fsdp2/SKILL.md
Commit: 773a529

Reviewed: about 13 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.