# ByteDance-Seed/Triton-distributed

Distributed Compiler and Optimized Parallel Kernels

Repository: https://github.com/ByteDance-Seed/Triton-distributed
Canonical: https://ross.abutalabs.com/products/triton-distributed
Homepage: https://triton-distributed.readthedocs.io/en/latest/
Language: Python
License: MIT
License Family: permissive
Last push: 2026-08-12T09:46:35+00:00

## Health v2 (maintenance only)
Score: 63/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 97, release rhythm 35, longevity 37
- inputs: {"age_days": 518, "days_push": 21, "days_rel": 355, "gap_med": 31.0, "n_releases_24m": 3}
- flags: prerelease_only
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1526, forks 167 (observed 2026-08-28T04:04:58.365279+00:00)

## What it is
Triton-distributed is a distributed compiler built on OpenAI Triton for computation-communication overlapping on multi-GPU systems. It lets developers write high-performance distributed kernels comparable to hand-optimized libraries like Distributed-GEMM and FLUX, targeting Nvidia and AMD GPUs.

## Use cases
- write distributed GPU kernels with computation-communication overlap
- implement expert parallelism for MoE models
- build low-latency all-to-all communication kernels
- optimize GEMM operations across multiple GPUs
- profile intra-kernel performance on GPUs
- port distributed kernel compilation to custom hardware

## When to choose
- you need custom distributed kernels beyond what standard Triton offers
- you are training or serving large models with expert parallelism
- you want Python-level kernel development instead of CUDA/CUTLASS
- you need computation and communication to overlap in a single kernel

## When to avoid
- you only need single-GPU kernels (plain Triton suffices)
- you want a turnkey distributed training framework rather than a kernel compiler
- you target CPUs or non-GPU accelerators without porting effort

## Facets
- artifact type: library
- maturity: active
- function: compiler, gpu-computing, machine-learning, llm-training, benchmarking
- domain: gpu-computing, microservices, deep-learning, large-language-models, developer-tools
- platform: python
- tags: triton, distributed-kernels, computation-communication-overlap, moe, expert-parallelism, nvidia-gpu, amd-gpu, gpu, linux

## Member repositories
- ByteDance-Seed/Triton-distributed (main) score 63

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:58.365279+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 2026-08-30T04:31:35.335649+00:00, confidence not recorded.
  - readme: https://github.com/ByteDance-Seed/Triton-distributed (fetched 2026-08-28T04:04:58.365279+00:00, sha 15a5df773896)
- Data as of 2026-08-30T08:39:29.467469+00:00.