M 2 XFP: A Metadata-Augmented Microscaling Data Format for Efficient Low-bit Quantization
Open Access
- 22 March 2026
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 1151-1167
Abstract
No abstract availableAll Related Versions
Funding Information
- National Natural Science Foundation of China (62222210)
- National Natural Science Foundation of China (62532006)
- National Natural Science Foundation of China (62502305)
- Shanghai Qi Zhi Institute Innovation Program (SQZ202316)
- Natural Science Foundation of Shanghai grants (25ZR1402275)
This publication has 24 references indexed in Scilit:
- BBAL: A Bidirectional Block Floating Point-Based Quantisation Accelerator for Large Language Models2025
- An Empirical Study of Microscaling Formats for Low-Precision LLM Training2025
- BitMoD: Bit-serial Mixture-of-Datatype LLM Acceleration2025
- VQ-LLM: High-performance Code Generation for Vector Quantization Augmented LLM Inference2025
- Anda: Unlocking Efficient LLM Inference with a Variable-Length Grouped Activation Data Format2025
- M-ANT: Efficient Low-bit Group Quantization for LLMs via Mathematically Adaptive Numerical Type2025
- AMD Instinct MI300X Generative AI Accelerator and Platform Architecture2024
- Inside Maia 1002024
- ANT: Exploiting Adaptive Numerical Data Type for Low-bit Deep Neural Network Quantization2022
- CACTI 7ACM Transactions on Architecture and Code Optimization, 2017