Distributed Training · 2023

FP8-LM / FP8 Training

Houwen Peng, Kan Wu, Han Hu, Peng Cheng, Paulius Micikevicius

FP8-LM shows large language models can be trained with 8-bit floating point for gradients, optimizer state, and communication by adding per-tensor scaling and precision decoupling, reducing memory and communication versus 16-bit training without loss degradation.

Editorial record

Plain-language summary

The framework extends mixed-precision training so that not just matrix multiplies but also gradients, the optimizer's moment states, and distributed communication use FP8, guarded by automatic per-tensor scaling to keep values inside FP8's narrow dynamic range and by decoupling precision across components to avoid underflow. Applied to GPT-scale models, it matched the accuracy of BF16 training while cutting memory footprint and communication volume and raising throughput. This pushed practical LLM training below 16 bits across more of the pipeline than earlier FP8 schemes, which had limited FP8 to the matmuls.

Knowledge graph

Relationships

Descendants

Source record

Provenance

Record ID
P-138
Record created
2026-07-13
Last reviewed
2026-07-14
Record version
2

Citation caveat: Citation metadata is approximate and marked unverified in the source dataset.