deepspeed

Expert guidance for distributed training with DeepSpeed - ZeRO optimization stages, pipeline parallelism, FP16/BF16/FP8, 1-bit Adam, sparse attention

Quality

47%

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 ./skills/distributed-training/deepspeed/SKILL.md

Quality

Content

25%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.

The content is essentially a raw documentation dump: enormous inlined text walls with a few buried executable examples and no clear workflow or progressive disclosure. It fails the token-efficiency and navigation goals of a skill body and should be restructured into a lean overview that points into the reference files.

Suggestions

Replace the inlined 'Pattern' walls with a concise overview that links into the existing reference files, moving the full tutorials and config reference out of SKILL.md.

Curate the quick-reference to keep only complete, executable code snippets and remove the fragmented single-token code blocks (e.g. 'libaio', '80', '32').

Add an explicit sequenced workflow with validation checkpoints (e.g. write config -> launch with deepspeed -> inspect wall_clock_breakdown/monitor -> validate convergence) instead of generic 'Working with This Skill' boilerplate.

Dimension	Reasoning	Score
Conciseness	The body is severely verbose: it inlines entire documentation pages (DeepNVMe tutorial, full feature overview, complete config JSON reference) as 'Pattern' walls of text, with repeated 'Updated: November 5, 2025 Previous Next' boilerplate and duplicated MoE/MoS content, heavily padding the context.	1 / 5
Actionability	It contains some genuinely executable, copy-paste-ready examples (e.g. FlopsProfiler training workflow, get_model_profile, aio_handle creation), but these are buried in prose and interleaved with many meaningless single-token code blocks ('libaio', '80', '32', '--mos'), leaving the overall guidance incomplete and noisy.	3 / 5
Workflow Clarity	There is no coherent sequenced workflow for using the skill; sections like 'When to Use' and generic 'Working with This Skill' boilerplate give only a rough, gap-filled structure with no validation checkpoints.	2 / 5
Progressive Disclosure	Although reference files exist and are listed, the body massively inlines content (full tutorials and the entire config reference) that clearly belongs in those separate files, and the reference descriptions are generic ('08.md - 08 documentation'), making navigation poor.	2 / 5
	Total	8 / 20 Passed

Description

70%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 specific and distinctive for the DeepSpeed niche with good trigger-term coverage, but it omits any explicit 'when to use' guidance, which caps completeness. Adding a 'Use when...' clause would lift it to a top score.

Suggestions

Append an explicit 'Use when...' clause, e.g. 'Use when setting up distributed training, configuring ZeRO stages, or debugging DeepSpeed performance.'

Reframe the feature list as concrete actions (e.g. 'Configure ZeRO stages', 'Enable FP16/BF16/FP8 mixed precision', 'Tune 1-bit Adam') to strengthen specificity.

Add common user phrasings/synonyms such as 'deepspeed config', 'ds_config.json', or 'ZeRO-3' to broaden trigger-term coverage.

Dimension	Reasoning	Score
Specificity	The description lists several specific features ('ZeRO optimization stages, pipeline parallelism, FP16/BF16/FP8, 1-bit Adam, sparse attention') but frames them as a feature list rather than concrete actions, leaving minor coverage gaps.	4 / 5
Completeness	It clearly states the 'what' (expert guidance for distributed training with DeepSpeed and named features) but provides no 'Use when...' clause or equivalent explicit trigger guidance, capping completeness at 3 per the rubric.	3 / 5
Trigger Term Quality	It covers strong natural terms users would say ('distributed training', 'DeepSpeed', 'ZeRO', 'pipeline parallelism', 'mixed precision', 'sparse attention') but omits a few common variations like config/JSON references.	4 / 5
Distinctiveness Conflict Risk	It targets a clear niche (DeepSpeed) with named, distinctive triggers (ZeRO, 1-bit Adam, FP8) that are unlikely to fire for unrelated skills, giving minimal conflict risk.	5 / 5
	Total	16 / 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: OpenLAIR/dr-claw
Path: skills/distributed-training/deepspeed/SKILL.md
Commit: 4b35756

Reviewed: 5 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.