huggingface-accelerate

Simplest distributed training API. 4 lines to add distributed support to any PyTorch script. Unified API for DeepSpeed/FSDP/Megatron/DDP. Automatic device placement, mixed precision (FP16/BF16/FP8). Interactive config, single launch command. HuggingFace ecosystem standard.

Quality

77%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Low

Low-risk findings worth noting

Fix and improve this skill with Tessl

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

Quality

Content

72%

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

The body is highly actionable and well-structured with verified progressive-disclosure references, but it repeats the core example and omits validation checkpoints for distributed/checkpoint workflows.

Suggestions

Remove the duplicate full conversion example in Workflow 1, instead referencing the Quick start snippet, to tighten conciseness.

Add explicit validation checkpoints for risky distributed operations, e.g. after `accelerate save_state` confirm the checkpoint exists on all ranks before proceeding, and after launch verify loss is non-NaN.

Consolidate the mixed-precision/DeepSpeed/FSDP plugin construction, which repeats the same `accelerator.prepare(...)` boilerplate, into a single shared pattern reference.

Dimension	Reasoning	Score
Conciseness	Mostly lean and code-dense with no beginner preamble, but the 4-line conversion example is duplicated between Quick start and Workflow 1, and lines like "Same code as before!" pad the body. Not a 3 because the repetition wastes tokens; not a 1 because it largely assumes Claude's competence.	2 / 3
Actionability	Fully executable, copy-paste-ready code, launch commands with concrete flags, and a complete deepspeed_config.json. Not below 3 because the guidance is specific and runnable rather than pseudocode.	3 / 3
Workflow Clarity	Workflows are clearly sequenced, but checkpointing, save_state, and distributed launches lack validation/verification checkpoints (no confirm-saved, check-loss, or verify-it-ran steps). Per the rubric, missing feedback loops for batch/distributed ops caps this at 2; not a 1 because the sequences are otherwise explicit.	2 / 3
Progressive Disclosure	SKILL.md is a concise overview with clearly signaled one-level-deep links to three real reference files (megatron-integration.md, custom-plugins.md, performance.md), all present in ./references/. Not below 3 because content is appropriately split and easy to navigate.	3 / 3
	Total	10 / 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, distinct, and rich in natural trigger terms, but omits an explicit "Use when..." clause, so it answers what the skill does without clearly stating when to invoke it.

Suggestions

Add an explicit "Use when..." clause, e.g. "Use when adding distributed/multi-GPU/multi-node support to a PyTorch script, or when switching between DDP, DeepSpeed, FSDP, and Megatron."

Trim the trailing "HuggingFace ecosystem standard." tagline, which is marketing fluff rather than a capability or trigger.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions: "4 lines to add distributed support", "Automatic device placement, mixed precision (FP16/BF16/FP8)", "Interactive config, single launch command". Not below 3 since these are specific and verifiable rather than vague.	3 / 3
Completeness	Answers "what" thoroughly but lacks any explicit "Use when..." / trigger clause for when Claude should invoke it. Per the rubric, a missing trigger clause caps completeness at 2; not a 3 because the "when" is only implied, never stated.	2 / 3
Trigger Term Quality	Contains natural terms a user would say: "distributed training", "PyTorch script", "mixed precision", and named engines "DeepSpeed/FSDP/Megatron/DDP". Good coverage; not below 3 as common variations are well represented.	3 / 3
Distinctiveness Conflict Risk	Clear niche as a distributed-training API in the HuggingFace ecosystem with named engine alternatives; unlikely to trigger for unrelated skills. Not below 3 because the scope is well-bounded.	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/accelerate/SKILL.md
Commit: 52845c3

Reviewed: about 10 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.