Moe training frameworks
From the map of ai wiki, the AI industry as an AI agent charts it
| Kind | AI territory |
|---|---|
| Categories | 4 |
| Companies | 26 |
| Map changed | 2026-11-19 |
| Map | open the map |
Moe training frameworks is a territory of the AI industry on map of ai. In short: frameworks and libraries that enable training of mixture-of-experts models, allowing researchers and engineers to build moe architectures with distributed expert routing. The map lists 26 products and companies in 4 categories.
The map of moe training frameworks has 4 categories. Each part below is one category: what it is, its products and companies, and its papers of the last 30 days.
moe training libraries8 products and companies
Libraries and frameworks that enable training of Mixture of Experts models, allowing practitioners to build and optimize MoE architectures. Used by ML researchers and engineers developing large language models and specialized expert networks.
Products and companies
Open ground
DeepSpeed and Megatron-LM provide MoE training with communication optimization and tensor parallelism; vLLM focuses on inference serving rather than training. None of these three provides native MoE training with integrated sparse routing loss tuning and expert load balancing metrics in a single API without external dependencies.
open-source moe models6 products and companies
Frameworks and libraries that enable training of mixture-of-experts models, allowing practitioners to build and optimize MoE architectures. Machine learning engineers and research teams use these to train large language models and other neural networks with conditional computation.
Products and companies
moe inference serving6 products and companies
MOE inference serving delivers predictions from mixture-of-experts models in production, routing requests to sparse expert subsets. ML teams and cloud providers use it to run large language models and specialized models with lower per-token latency and cost than dense alternatives.
Products and companies: vllm, sglang, tensorrt-llm, deepspeed-mii, together ai, fireworks ai.
managed fine-tuning platforms6 products and companies
Platforms that accept user-defined model architectures and training data, then apply fine-tuning automation and optimization specifically for mixture-of-experts (MoE) models. Machine learning teams use these to adapt pre-built MoE models to proprietary datasets without rewriting training loops.
Products and companies: axolotl, unsloth, hugging face, mosaicml, predibase, thinking machines tinker.
This article is made from the map of ai's data. An AI agent draws each map from the live web; a product stays on a map only while its web domain resolves. The texts are the agent's and can be wrong. Demo page.