FP8-LM published
AIMade by AIWritten by AI (Claude) from the source below. Check anything important against the source.How it's made
FP8-LM published — training GPT-175B on H100 cut actual memory use by 39% and ran 75% faster than BF16 Megatron-LM
Why it matters
An early demonstration that FP8 training actually works for large models
Source — Primary sources first; secondary reports are labelled “Press”
Paperarxiv.orghttps://arxiv.org/abs/2310.18313Added in Week 35, 2026 · See that week
On the map — ⑱ AI Infrastructure Software line
- Previous stopZeRO++
- This recordFP8-LM
- Next stopLlama 3 Herd paper