# volcengine/veScale

Byted PyTorch Distributed for Hyperscale Training of LLMs and RLs

Repository: https://github.com/volcengine/veScale
Canonical: https://ross.abutalabs.com/products/vescale
Language: Python
License: Apache-2.0
License Family: permissive
Topics: llm-training, pytorch
Last push: 2026-03-03T22:19:43+00:00

## Health v2 (maintenance only)
Score: 57/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 70, release rhythm 35, longevity 65
- inputs: {"age_days": 919, "days_push": 183, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1036, forks 64 (observed 2026-08-28T04:03:19.226351+00:00)

## What it is
veScale is a PyTorch distributed training library from ByteDance for hyperscale training of large language models and reinforcement learning workloads. It provides eager-mode SPMD tensor programming and a flexible, high-performance FSDP implementation.

## Use cases
- train large language models across many GPUs
- run distributed FSDP training at scale
- train RL models with PyTorch distributed
- shard tensors with ragged sharding in DTensor
- scale PyTorch training to hyperscale clusters

## When to choose
- you need a PyTorch-native distributed LLM training library
- you want a flexible high-performance alternative to stock FSDP
- you train LLMs or RL workloads on large GPU clusters

## When to avoid
- you need a stable, feature-complete framework today (the repo is mid-migration with old code moved to legacy/)
- you train small models on a single GPU
- you need a framework-agnostic solution outside PyTorch

## Facets
- artifact type: library
- maturity: experimental
- function: llm-training, machine-learning, deep-learning
- domain: machine-learning, deep-learning, large-language-models, gpu-computing
- platform: python
- tags: pytorch, distributed-training, fsdp, reinforcement-learning, spmd, dtensor, gpu, linux

## Member repositories
- volcengine/veScale (main) score 57

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:19.226351+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T07:04:43.799734+00:00, confidence not recorded.
  - readme: https://github.com/volcengine/veScale (fetched 2026-08-28T04:03:19.226351+00:00, sha 503665393282)
  - registry_pypi: https://pypi.org/pypi/vescale/json (fetched 2026-08-29T13:05:46.734494+00:00, sha a2dd1259dade)
- Data as of 2026-08-30T08:39:29.467469+00:00.