# InfiX.ai's miniTE speeds learner updates 1.33–1.37× over BF16 on 8×B300

> Original title: 🚀 miniTE: 1.33–1.37× BF16 speed for learner updates on 8×B300.

- Source: LMSYS Org
- Published: 2026-10-27T18:28:36.000Z
- Bad Influence: https://badinfluence.news/items/kxl7ouecexz4ne8z1u0tuicjn
- Original: https://x.com/lmsysorg/status/2115148675546787876

## Summary

InfiX.ai's miniTE is a low-precision training framework that reduces training-sampling mismatch and speeds up NVFP4 computation. LMSYS Org reports 1.33–1.37× BF16 speed for learner updates on 8×B300 hardware. The framework diagnoses mismatch from quantization through token probabilities, controls risky updates with TRIAGE, and cuts compute overhead with grouped GEMM, contiguous buffers, and fusion.

## Post

🚀 miniTE: 1.33–1.37× BF16 speed for learner updates on 8×B300.

http://InfiX.ai’s miniTE is a low-precision training framework that reduces training–sampling mismatch and speeds up NVFP4 computation.

Highlights:
- Diagnose mismatch from quantization to token probabilities.
- Control risky updates with TRIAGE.
- Reduce compute overhead with grouped GEMM, contiguous buffers, and fusion.

Read the full report below 👇

![](https://badinfluence.news/api/img-proxy?u=https%3A%2F%2Fpbs.twimg.com%2Fmedia%2FHVqELIAacAAD4m5.jpg&mode=thumb&exp=1793404800&sig=93b8705036d0043b)