distributed-offline-packing

Official

Distribute and pack large SFT data across nodes.

AuthorEvolvingLMMs-Lab
Version1.0.0
Installs0

System Documentation

What problem does it solve?

Use this skill to orchestrate the multi-node packing of enormous SFT JSONL datasets into Energon WebDataset shards, enabling scalable SFT data preparation with padding-free shards and a consolidated metadataset.

Core Features & Use Cases

  • End-to-end distributed packing across multiple nodes using offline_packing/auto_pipe.sh to partition work, tokenize/prompts, bin samples, and generate webdataset shards.
  • Per-node Metadataset assembly that aggregates node outputs into a single logical dataset, suitable for large-scale SFT pipelines.
  • Prerequisites and guardrails including shared NFS, consistent Docker image, and script stages s1–s4 to ensure deterministic reproducibility.

Quick Start

Prepare an N-node environment with a shared NFS, mount the repository, and run offline_packing/auto_pipe.sh on your Part JSONL to produce Energon WebDataset shards.

Dependency Matrix

Required Modules

None required

Components

Standard package

💻 Claude Code Installation

Recommended: Let Claude install automatically. Simply copy and paste the text below to Claude Code.

Please help me install this Skill:
Name: distributed-offline-packing
Download link: https://github.com/EvolvingLMMs-Lab/LLaVA-OneVision-2/archive/main.zip#distributed-offline-packing

Please download this .zip file, extract it, and install it in the .claude/skills/ directory.
View Source Repository

Agent Skills Search Helper

Install a tiny helper to your Agent, search and equip skill from 471,000+ vetted skills library on demand.