quantdiff/CITATION.cff at main · federicobrancasi/quantdiff · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
cff-version: 1.2.0
message: "If you use this software, please cite it as below."
type: software
title: "QuantDiff: Efficient Mixed-Precision Quantization for Stable Diffusion via Sensitivity-Driven Optimization"
authors:
  - family-names: "Brancasi"
    given-names: "Federico"
    affiliation: "CERN"
  - family-names: "Pierini"
    given-names: "Maurizio"
    affiliation: "CERN"
  - family-names: "Segal"
    given-names: "Shai"
    affiliation: "CEVA"
  - family-names: "Janco"
    given-names: "Roy"
    affiliation: "CEVA"
  - family-names: "Klempner"
    given-names: "Anat"
    affiliation: "CEVA"
  - family-names: "Radiano"
    given-names: "Eyal"
    affiliation: "CEVA"
repository-code: "https://github.com/federicobrancasi/quantdiff"
license: Apache-2.0
keywords:
  - quantization
  - stable-diffusion
  - mixed-precision
  - deep-learning
  - optimization
abstract: >-
  QuantDiff is an automated framework for mixed-precision quantization of
  Stable Diffusion models. It uses a three-phase approach: computational cost
  analysis (BOPs), layer sensitivity measurement (FID/CLIP metrics), and
  greedy Bang-for-Buck optimization to allocate bit-widths intelligently.
  The framework achieves significant computational cost reduction while
  preserving image generation quality.