AI Paper Atlas

Research map / Architectures and efficiency / Quantization

Is Finer Better? The Limits of Microscaling Formats in Large Language Models

Andrea Fasoli, Monodeep Kar, Chi-Chun Liu, Swagath Venkataramani, Viji Srinivasan, Leland Chang, Naigang Wang

ICLR 2026 · Architectures and efficiency › Quantization · approach: KV cache quantization · 7 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
use FP8 UE5M3 scales instead of UE4M3 (ablation)FP4 microscaling, granite-3.3-8b · Wikitext2 perplexity (FP4, block size 8)perplexity: 7.43 -> 5.04Improved
use FP8 UE5M3 scales instead of UE4M3 (ablation)FP4 microscaling, bamba-9b-v2 · Wikitext2 perplexity (FP4, block size 8)perplexity: 21.25 -> 6.53Improved
use FP8 UE5M3 scales instead of UE4M3 (ablation)FP4 microscaling, granite-3.3-8b · GSM8K (FP4, block size 8)accuracy: 32.37 -> 56.17Improved
use FP8 UE5M3 scales instead of UE4M3 (ablation)FP4 microscaling, bamba-9b-v2 · GSM8K (FP4, block size 8)accuracy: 2.65 -> 39.42Improved
use FP8 UE5M3 scales instead of UE4M3 (ablation)FP4 microscaling, nemotron-nano-9b-v2 · GSM8K (FP4, block size 8)accuracy: 72.71 -> 77.71Improved
use FP8 UE5M3 scales instead of UE4M3 (ablation)FP4 microscaling, llama-3.1-8b · Wikitext2 perplexity (FP4, block size 8)perplexity: 7.23 -> 6.79Improved
add dynamic per-tensor scaling to UE4M3 scales (ablation)FP4 microscaling, granite-3.3-8b · GSM8K (FP4, block size 8)accuracy: 32.37 -> 44.88Improved
add dynamic per-tensor scaling to UE4M3 scales (ablation)FP4 microscaling, bamba-9b-v2 · Wikitext2 perplexity (FP4, block size 8)perplexity: 21.25 -> 6.53Improved
use UE5M3 without per-tensor scaling instead of UE4M3 with per-tensor scaling (ablation)FP4 microscaling, granite-3.3-8b · GSM8K (FP4, block size 8)accuracy: 44.88 -> 56.17Improved
use UE5M3 without per-tensor scaling instead of UE4M3 with per-tensor scaling (ablation)FP4 microscaling, llama-3.1-8b · GSM8K (FP4, block size 8)accuracy: 43.21 -> 42.15No clear effect
use UE5M3 without per-tensor scaling instead of UE4M3 with per-tensor scaling (ablation)FP4 microscaling, bamba-9b-v2 · GSM8K (FP4, block size 8)accuracy: 40.41 -> 39.42No clear effect
use UE5M3 without per-tensor scaling instead of UE4M3 with per-tensor scaling (ablation)FP4 microscaling, nemotron-nano-9b-v2 · GSM8K (FP4, block size 8)accuracy: 78.92 -> 77.71No clear effect

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