No candidate page describes a sparse autoencoder training tool directly. The closest fits are infrastructure and ML platforms that can run training jobs: Modal provides cloud infrastructure for training and serving AI models at scale 10, NVIDIA's Brev.dev console offers instances and includes tao-train-mask-auto-encoder for masked auto-encoder pretraining 515, and Fast's fastai provides deep learning training tooling and PyTorch libraries 2.

A question back: Which platform or hardware do you plan to train on (for example a cloud GPU provider, your own GPUs, or a specific framework such as PyTorch)?

1. 1

   Modalmodal.comverified and signed

   Cloud infrastructure for training and serving AI models at scale, including multi-node GPU training runs for custom models 10.

   WebsiteDocsPricingGet started

   +16 community tools built for it

   for agents: serverless · infrastructure · cloud

2. 2

   Brev.devnvidia.comverified and signed

   Instances console for running GPU workloads; its indexed action tao-train-mask-auto-encoder trains masked auto-encoders for self-supervised visual pretraining 15.

   WebsiteDocsPricingGet started

   +81 community tools built for it

   for agents: tao-train-mask-auto-encoder · tao-train-sparse4d · nemo-automodel-recipe-development

3. 3

   fastaifast.aiverified and signed

   Deep learning library for PyTorch used to build and train neural networks, a practical base for writing a custom sparse autoencoder 2.

   WebsiteDocsGet started

   +460 community tools built for it

   for agents: python · github · api

4. 4

   Mistralmistral.aiverified and signed

   Open-weight models you can customize with Forge and run on-prem or in your VPC, useful as a base model to inspect or train against 3.

   WebsiteDocsPricingGet started

   +297 community tools built for it

   for agents: coding-assistant · llm · developer-tools