# vLLM-Omni runs MAGI-2 Preview on Moore Threads MUSA, cutting step time from 15.6 s to 565 ms

> Original title: MAGI-2 Preview on vLLM-Omni: Bringing a 114B Audio-Video MoE DiT to Moore Threads MUSA with torchada

- Source: vLLM Blog
- Published: 2026-10-12T00:00:00.000Z
- Bad Influence: https://badinfluence.news/items/f7q8rnpkwoeghdnfukzt5o226
- Original: https://vllm.ai/blog/2026-10-12-magi2-preview-musa

## Summary

The vLLM team reports running MAGI-2 Preview, a 114B audio-video MoE DiT, on Moore Threads MTT S5000 GPUs using torchada to map CUDA-style calls to MUSA. Per-step time fell from 15.6 s in the first working run to 565 ms with all listed pull requests applied on 8 GPUs at 272p and 100 steps. The post pairs each change with an A/B measurement and an output hash check, and states that several pull requests remain open.

## Why it matters

The post traces a step-by-step optimization path with paired measurements and bit-identical checks, useful for engineers porting CUDA-first models to other accelerators.