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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| train in BF16 instead of FP32 | DistilBERT fine-tuning · AG News | throughput (data/s): 1263.3 -> 2944.8 | Improved |
| train in BF16 instead of FP32 | DistilBERT fine-tuning · AG News | accuracy (%): 93.45 -> 93.49 | No clear effect |
| train in FP8 instead of FP32 | DistilBERT fine-tuning · AG News | peak memory (GB): 3.28 -> 2.24 | Improved |
| train in FP8 instead of FP32 | DistilBERT fine-tuning · IMDB | throughput (data/s): 476.2 -> 928.7 | Improved |
| train in FP8 instead of BF16 | DistilBERT fine-tuning · IMDB | throughput (data/s): 1005.3 -> 928.7 | Hurt |
| train in FP8 instead of FP32 | DistilBERT fine-tuning · IMDB | accuracy (%): 91.18 -> 91.38 | No clear effect |
| use simulated INT8 instead of FP32 | DistilBERT fine-tuning · AG News | accuracy (%): 93.45 -> 91.36 | Hurt |
| use simulated INT8 instead of FP32 | DistilBERT fine-tuning · IMDB | accuracy (%): 91.18 -> 89.19 | Hurt |
| use simulated FP4 instead of FP32 | DistilBERT fine-tuning · AG News | accuracy (%): 93.45 -> 89.80 | Hurt |
| use simulated FP4 instead of FP32 | DistilBERT fine-tuning · IMDB | accuracy (%): 91.18 -> 80.44 | Hurt |
| train in FP16 instead of FP32 | DistilBERT fine-tuning · IMDB | peak memory (GB): 2.06 -> 1.71 | Improved |
Closest papers by meaning
- Optimizing Large Language Model Training Using FP4 Quantization (ICML 2025 · similarity 0.92)
- Any-Precision LLM: Low-Cost Deployment of Multiple, Different-Sized LLMs (ICML 2024 · similarity 0.92)
- Benchmarking Optimizers for Large Language Model Pretraining (NeurIPS 2026 · similarity 0.92)
- QLLM: Accurate and Efficient Low-Bitwidth Quantization for Large Language Models (ICLR 2024 · similarity 0.91)
- Norm Tweaking: High-Performance Low-Bit Quantization of Large Language Models (AAAI 2024 · similarity 0.91)
- Exploring Post-training Quantization in LLMs from Comprehensive Study to Low Rank Compensation (AAAI 2024 · similarity 0.91)
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.
