rwkv-architecture

RNN+Transformer hybrid with O(n) inference. Linear time, infinite context, no KV cache. Train like GPT (parallel), infer like RNN (sequential). Linux Foundation AI project. Production at Windows, Office, NeMo. RWKV-7 (March 2025). Models up to 14B parameters.

Quality

61%

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/rwkv/SKILL.md

Quality

Content

72%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 body is highly actionable with executable, copy-paste-ready code across the main workflows and good progressive disclosure into real reference files. It loses points on conciseness from padded comparison prose and inline time-sensitive info, and on workflow clarity because the batch fine-tuning workflow omits validation checkpoints.

Suggestions

Add explicit validation/verification steps to the fine-tuning workflow (e.g., validate dataset, checkpoint inspection, eval-passes-before-proceed).

Trim the Transformer-vs-RWKV comparison prose and move per-token computation tallies into a concise table or the architecture reference.

Move time-sensitive version/date info (RWKV-7 March 2025, arxiv dates) into a dedicated version-history or references section rather than inline.

Dimension	Reasoning	Score
Conciseness	Mostly efficient with concrete code, but the Transformer-comparison prose ('1,000,000× more efficient!', per-token computation tallies) is padding, and time-sensitive version/date info (RWKV-7 March 2025, inline arxiv dates) is not isolated in a deprecated section.	3 / 5
Actionability	Provides copy-paste-ready executable code for installation, GPT/RNN modes, streaming generation, long-context streaming, and fine-tuning, covering the common cases with specific commands and full snippets.	5 / 5
Workflow Clarity	Workflows are clearly labeled, but the batch fine-tuning workflow (8-GPU DeepSpeed training) lacks validation/verification checkpoints, capping workflow clarity at 3 for batch operations per the guidelines.	3 / 5
Progressive Disclosure	Clear overview with well-signaled, one-level-deep references to real bundle files (architecture-details.md, state-management.md, rwkv7.md), all verified to exist and appropriately split out.	5 / 5
	Total	16 / 20 Passed

Description

51%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 technically rich and distinct but reads as a feature list rather than trigger-oriented guidance; it omits any explicit 'Use when...' clause, capping completeness. Trigger terms lean toward jargon over natural user phrasing.

Suggestions

Add an explicit 'Use when...' clause naming natural triggers (e.g., 'Use when working with RWKV models, linear-complexity RNNs, or infinite-context inference').

Replace some technical jargon with user-facing terms and synonyms (e.g., add 'long-context language model', 'constant-memory inference') to improve trigger quality.

Lead with the concrete actions the skill performs (load, run, fine-tune, compare RWKV models) rather than architectural attributes.

Dimension	Reasoning	Score
Specificity	Names the domain and several concrete technical specifics ('RNN+Transformer hybrid with O(n) inference', 'Train like GPT (parallel), infer like RNN (sequential)', 'Models up to 14B parameters'), but these are attributes rather than discrete user-facing actions.	4 / 5
Completeness	Clearly states 'what' the skill covers, but there is no 'Use when...' or equivalent explicit trigger clause, capping completeness at 3 per the guidelines.	3 / 5
Trigger Term Quality	Terms are mostly technical jargon ('O(n) inference', 'no KV cache', 'receptance weighted key value') rather than natural phrases a user would say; missing common user-facing synonyms.	2 / 5
Distinctiveness Conflict Risk	The RWKV niche and named production deployments are mostly distinct, with only minor overlap risk against general RNN/Transformer or efficient-inference skills.	4 / 5
	Total	13 / 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: synthetic-sciences/openscience
Path: backend/cli/skills/ml-training/rwkv/SKILL.md
Commit: 3a6c3a9

Reviewed: 1 day 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.