# open-gigaai/giga-train

GigaTrain: An Efficient and Scalable Training Framework for AI Models

Repository: https://github.com/open-gigaai/giga-train
Canonical: https://ross.abutalabs.com/products/giga-train
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-08-25T12:39:58+00:00

## Health v2 (maintenance only)
Score: 62/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 99, release rhythm 35, longevity 24
- inputs: {"age_days": 341, "days_push": 8, "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 1072, forks 94 (observed 2026-08-28T04:03:28.393032+00:00)

## What it is
GigaTrain is an efficient and scalable Python training framework for large AI models, supporting distributed multi-GPU/multi-node execution via DeepSpeed ZeRO, FSDP, and DDP. It provides YAML/JSON configuration, mixed precision (FP16/BF16/FP8), gradient checkpointing, EMA, and built-in logging and checkpointing.

## Use cases
- fine-tune large AI models on multiple GPUs
- train models across multiple nodes with DeepSpeed ZeRO
- run reproducible training experiments from YAML configs
- fine-tune the Wan video model
- resume long training runs from checkpoints
- train with FP8 or BF16 mixed precision to save memory

## When to choose
- you need scalable multi-GPU/multi-node training with ZeRO or FSDP
- you want a lightweight framework that handles backprop, logging, checkpointing, and EMA for you
- you prefer config-driven, reproducible training pipelines
- you are fine-tuning large generative models like Wan

## When to avoid
- you only need inference or model serving rather than training
- you need a framework with first-class support for a specific niche model architecture not covered by its registry
- you prefer writing fully custom training loops from scratch

## Facets
- artifact type: framework
- maturity: active
- function: llm-training, machine-learning, deep-learning, gpu-computing, monitoring, configuration-management
- domain: machine-learning, deep-learning, large-language-models, artificial-intelligence, gpu-computing, developer-tools
- platform: python, cross-platform
- tags: distributed-training, deepspeed, fsdp, mixed-precision, fine-tuning, training-framework, multi-gpu, checkpointing, linux, gpu, docker

## Member repositories
- open-gigaai/giga-train (main) score 62

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:28.393032+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 2026-08-30T06:53:42.636058+00:00, confidence not recorded.
  - readme: https://github.com/open-gigaai/giga-train (fetched 2026-08-28T04:03:28.393032+00:00, sha c5fa26ed6f73)
  - registry_pypi: https://pypi.org/pypi/giga-train/json (fetched 2026-08-29T12:56:02.762409+00:00, sha cbd6ec171fd1)
- Data as of 2026-08-30T08:39:29.467469+00:00.