MXFP8-E4M3 with round-up scale computation matches BF16 pre-training accuracy for LLMs up to 8B parameters, enabling 2x faster training on Blackwell.
IEEE Std 754-2008 , pages 1–70, 2008
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Recipes for Pre-training LLMs with MXFP8
MXFP8-E4M3 with round-up scale computation matches BF16 pre-training accuracy for LLMs up to 8B parameters, enabling 2x faster training on Blackwell.