AI Paper Atlas

Research map / Architectures and efficiency / Quantization

Low-Precision Training of Large Language Models: Methods, Challenges, and Opportunities

Zhiwei Hao, Jianyuan Guo, Li Shen, Yong Luo, Han Hu, Guoxia Wang, Dianhai Yu, Yonggang Wen, Dacheng Tao

TPAMI 2026 · Architectures and efficiency › Quantization · approach: Post-training quantization of LLMs · 16 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
train in BF16 instead of FP32DistilBERT fine-tuning · AG Newsthroughput (data/s): 1263.3 -> 2944.8Improved
train in BF16 instead of FP32DistilBERT fine-tuning · AG Newsaccuracy (%): 93.45 -> 93.49No clear effect
train in FP8 instead of FP32DistilBERT fine-tuning · AG Newspeak memory (GB): 3.28 -> 2.24Improved
train in FP8 instead of FP32DistilBERT fine-tuning · IMDBthroughput (data/s): 476.2 -> 928.7Improved
train in FP8 instead of BF16DistilBERT fine-tuning · IMDBthroughput (data/s): 1005.3 -> 928.7Hurt
train in FP8 instead of FP32DistilBERT fine-tuning · IMDBaccuracy (%): 91.18 -> 91.38No clear effect
use simulated INT8 instead of FP32DistilBERT fine-tuning · AG Newsaccuracy (%): 93.45 -> 91.36Hurt
use simulated INT8 instead of FP32DistilBERT fine-tuning · IMDBaccuracy (%): 91.18 -> 89.19Hurt
use simulated FP4 instead of FP32DistilBERT fine-tuning · AG Newsaccuracy (%): 93.45 -> 89.80Hurt
use simulated FP4 instead of FP32DistilBERT fine-tuning · IMDBaccuracy (%): 91.18 -> 80.44Hurt
train in FP16 instead of FP32DistilBERT fine-tuning · IMDBpeak memory (GB): 2.06 -> 1.71Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy