method · first seen 31 Oct 2026
4-bit weight quantization
Also: INT4
Compressing neural network weights to 4-bit precision to reduce memory and speed up inference.
Mentions per week
PapersCodeDiscussion
Earliest sources
- Integer-Exact Microscaling: Running MXFP4 Language Models on INT8 Matrix UnitsPapers · 31 Oct 2026 · Shinyoung Kang, Junhyung Ahn, Yujun Kim, Jungho Park
Latest sources
- Integer-Exact Microscaling: Running MXFP4 Language Models on INT8 Matrix UnitsPapers · 31 Oct 2026 · Shinyoung Kang, Junhyung Ahn, Yujun Kim, Jungho Park