Event ·Official first-hand

摩尔线程开源 Megatron-MUSA-Titan，支持万亿参数 MoE 训练

摩尔线程开源了 Megatron-MUSA-Titan 训练框架，这是其用于在 10000 卡 GPU 集群上进行万亿参数 MoE 预训练的内部训练栈。该框架采用块级 FP8（blockwise FP8）技术，并支持 slime 和 miles 两个 RL 训练工具。这意味着开发者可以在摩尔线程国产 GPU 上复用 NVIDIA Megatron 生态的大规模训练方案。

Current read

摩尔线程把自家万卡集群训万亿参数 MoE 的训练栈开源了，带块级 FP8，国产 GPU 上也能跑 Megatron 式大规模训练。

11 sources57° AI heatLast updated 2026/10/29 02:00:28

Evidence

11 sources

Primary sourcepandaily

2026/10/29 02:00:28View source

Related source 1IT之家

2026/10/28 10:16:40View source

Related source 2rohanpaul_ai

2026/10/27 16:41:33View source

Related source 3techcrunch

2026/10/28 15:00:00View source

Related source 4Artificial Analysis

2026/10/28 15:05:14View source

Related source 5vLLM

2026/10/28 15:26:59View source

Related source 6Decoder

2026/10/28 15:49:39View source

Related source 7kimmonismus

2026/10/28 20:45:10View source

Related source 8TestingCatalog

2026/10/28 09:58:40View source

Related source 9Hugging Face: Blog

2026/10/28 14:58:17View source

Related source 10Gorden Sun

2026/10/28 16:01:32View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story