CtrlK
BlogDocsLog inGet started
Tessl Logo

veomni-migrate-transformers-v5

Use this skill when adding or refreshing a patchgen-generated modeling file for a VeOmni model under its generated directory — GPU-only or GPU+NPU, dense or MoE, text-only / VLM / Omni-thinker+talker. Covers: creating GPU and NPU patchgen configs, using patchgen decorators (replace_class/override_method/replace_function/modify_init/add_post_import_block/drop_import_names), reusing sibling-model patches via name_map, handling MoE weight-loading (CheckpointTensorConverter + fused gate_up_proj layout), multimodal/VLM forward with Ulysses SP, excluding speech/vocoder subtrees in Omni models (talker/token2wav/DiT/BigVGAN), wiring __init__.py for the patchgen-generated classes, running codegen, and adding test cases. Trigger: 'port a model to patchgen', 'add patchgen for a model', 'transformers v5 migration', 'add NPU patchgen'. Do NOT edit files under generated/ manually — always regenerate via patchgen.

71

Quality

88%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

SecuritybySnyk

Passed

No findings from the security scan

SKILL.md
Quality
Evals
Security

Quality

Content

77%

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

The content is highly actionable with a clear phased workflow and validation checkpoints, but it is over-long and self-duplicating, and it keeps a large amount of material inline rather than splitting it into reference files.

Suggestions

Consolidate the "Common Pitfalls" section so each pitfall is stated once (in its phase or in pitfalls, not both) to cut the ~1000-line body and reduce token cost.

Move the patch-primitive table, the full per-category example walkthroughs, and the MoE converter layout details into a reference file under references/, keeping SKILL.md as a lean overview with one-level-deep pointers.

Consider extracting the long Phase 2 patch-set and Phase 3 converter recipes into separate referenced docs, signaling them clearly from the main body.

DimensionReasoningScore

Conciseness

The body is a ~1000-line document with substantial duplication: the "Common Pitfalls" section restates many points already covered in the phases (e.g. MoE layouts, converter round-trip safety, dummy_forward dtype, MoE _moe_implementation propagation). It is mostly actionable but could be tightened by de-duplicating pitfalls against the phase content.

2 / 3

Actionability

Provides concrete executable commands (patchgen regen, pytest invocations, uv sync), real code skeletons (PatchConfig, register patterns), and specific file/test paths, making guidance copy-paste ready.

3 / 3

Workflow Clarity

Clear 8-phase sequence with explicit "Validation" checkpoints at the end of most phases and a verify-diff/CI-drift loop in Phase 5 ("If errors: fix and re-validate", "patchgen --check"), satisfying the explicit-validation anchor.

3 / 3

Progressive Disclosure

References to docs and example directories are signaled one level deep, but no bundle files actually exist in the skill and the body itself is a near-monolithic ~1000 lines with content (e.g. the full Common Pitfalls catalog and patch primitive tables) that could live in separate reference files.

2 / 3

Total

10

/

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.

The description is comprehensive and well-structured, clearly stating capabilities, explicit triggers, and a sharp niche. It is dense but every clause adds concrete information rather than fluff.

DimensionReasoningScore

Specificity

Enumerates many concrete actions ("creating GPU and NPU patchgen configs", "using patchgen decorators (replace_class/override_method/replace_function/modify_init/add_post_import_block/drop_import_names)", "wiring __init__.py", "running codegen", "adding test cases"), matching the multiple-specific-actions anchor.

3 / 3

Completeness

Explicitly answers both what ("Covers: ...") and when ("Trigger: ...") with explicit triggers, satisfying the both-what-and-when anchor.

3 / 3

Trigger Term Quality

Includes natural trigger phrases users would actually say ("Trigger: 'port a model to patchgen', 'add patchgen for a model', 'transformers v5 migration', 'add NPU patchgen'") with good coverage.

3 / 3

Distinctiveness Conflict Risk

Highly specific niche (VeOmni patchgen-generated modeling for transformers v5) with distinct triggers, unlikely to conflict with other skills.

3 / 3

Total

12

/

12

Passed

Validation

87%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation14 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

skill_md_line_count

SKILL.md is long (1030 lines); consider splitting into references/ and linking

Warning

referenced_paths_exist

Referenced path issues: 1 missing, 1 deeper-than-1-level

Warning

Total

14

/

16

Passed

Repository
ByteDance-Seed/VeOmni
Reviewed

Table of Contents

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.