OpenTrain AI
No verified implementation yet

Unleashing Low-Bit Inference on Ascend NPUs: A Comprehensive Evaluation of HiFloat Formats

Pengxiang Zhao, Hui-Ling Zhen, Xing Li, Han Bao, Weizhe Lin +8 more

February 13, 2026arXiv: 2602.12635
0 repos~a few days to reproduce
arXiv PDF

Abstract

As LLMs scale, low-bit floating-point formats like MXFP and NVFP4 offer new opportunities for precision and efficiency. In this work, we evaluate HiFloat (HiF8 and HiF4), a family of formats tailored for Ascend NPUs. Through rigorous comparison across weight-activation and KV-cache tasks, we provide three key insights: (1) INT8 suits narrow-range data, while floating-point formats excel with high-variance data; (2) i...

Results & Benchmarks

TaskDatasetMetricValue
QuantizationHellaSwagPerplexity500
QuantizationMMLUPerplexity500
QuantizationARCPerplexity500
QuantizationMATHPerplexity500

Hardware Requirements

  • Expect multi-day setup/compute for meaningful reproduction based on current guidance.

Best Implementation

Maintained implementation evidence is not confirmed for this paper yet.

Use the Implementation Status and Reproduction Path sections below for the current action plan.

Reproduction Path

Follow this baseline workflow to decide if this paper is worth immediate prototyping.

  1. 1

    Use the paper and benchmark evidence to scope a baseline reproduction plan.

  2. 2

    Start from this likely method family: Quantization.

  3. 3

    Track assumptions and missing details in an experiment log before coding.

Time to first repro: a few daysEstimate is based on paper-only reproduction flow

Additional Implementations

No additional verified repositories beyond the primary recommendation.

Hugging Face Artifacts

No trustworthy direct or curated related Hugging Face artifacts were found yet.

Research Context