Story

arxiv_cs_lg ยท Nov 16, 2026 ยท paper

Source brief

CertiSum: Bitwise-Identical MXFP4 Training Across Platforms and Parallelism Layouts

arxiv.orgNov 16, 2026
original source linked

In brief

Training with MXFP4, a 4-bit microscaling format, is non-deterministic: identical inputs can produce different results across parallelism layouts and GPU vendors. These differences complicate debugging and training au...

Feed lens
eval

Continue reading

Read the original at arxiv.org โ†’Open in live feedRead that dayโ€™s brief

Earlier in this thread 4 items