ByteDance-SeedOfficial·9 Agent Skills Included

VeOmni

AI-assisted development toolkit for distributed multi-modal model training

Guides AI agents through building, debugging, and optimizing VeOmni, a distributed training framework for text, vision, audio, and omni models. Covers adding new models, writing fused kernels, migrating to transformers v5, profiling training speed, and fixing distributed training bugs. Removes guesswork with step-by-step protocols, code review gates, and pull request automation.

2,135|247

↗

npx skills add ByteDance-Seed/VeOmni --all -g -y

Available:

Tells the AI agent how to load project knowledge, follow core engineering principles, and route each task to the right skill such as debugging, model integration, profiling, or pull request creation.

All Skills in This Repository (9)

Pure Emerald Level Indicators

📦 In Repo

ByteDance-Seed

veomni-new-model

Integrate HuggingFace models into VeOmni with patches and configuration.

veomni-profile

Profile deep learning training runs by parsing Chrome traces and memory snapshots.

veomni-debug

Debug distributed training errors with structured investigation protocols.

veomni-migrate-transformers-v5

Migrate transformer model code to the v5 patchgen framework.

veomni-uv-update

Update uv, torch, and related dependencies across VeOmni configuration files and Docker images.

veomni-develop

Evaluate and refactor VeOmni components with impact analysis and safety rules.

create-pr

Automate pull request creation by analyzing git status and generating standardized draft files.

veomni-review

Analyze git diffs against constraint files to detect risks and violations.

veomni-new-op

Assists adding, testing and documenting new operators in the SSA framework.

Official

Advanced

Frequently Asked Questions

FAQPage Schema

How to install VeOmni skills?▼

Run `npx skills add ByteDance-Seed/VeOmni --all -g -y` in your terminal to install all skills in this suite globally.

What is VeOmni used for?▼

VeOmni is a distributed training framework for scaling text, vision, audio, and omni models across GPUs and NPUs. These skills help an AI agent develop, debug, and optimize that framework.

Can AI add a new model to VeOmni?▼

Yes. The veomni-new-model skill walks the agent through analyzing the HuggingFace model, creating patches, defining parallel plans, writing configs, and testing.

How to debug distributed training hangs with AI?▼

The veomni-debug skill gives the agent a structured protocol for root-cause analysis, hypothesis testing, and fixing issues like loss divergence, FSDP errors, and training hangs.

Do these skills work with Claude Code and Cursor?▼

Yes. All skills follow the open SKILL.md standard and are auto-discovered by Claude Code, Cursor, Codex, and other compatible agents.