mlfont-notosanssc: Range-Compact MLFont Slicing for Noto Sans SC (Simplified Chinese)

This is not an officially supported Google product.

Overview

Pretrained MLFont machine-learning-guided font slicing model and WOFF2 slice bundle for Noto Sans SC (Simplified Chinese) (NotoSansSC-Regular.otf, 30,759 Unicode codepoints partitioned into 102 range-compact slices).

• Paper: MLFont: Machine Learning-Driven Font Subsetting and Optimal Unicode-Range Slicing for Low-Latency Web Typography (Quan Wang and Yiran Mao)
• Open-Source Repository: https://github.com/wq2012/mlfont
• PyPI Package: https://pypi.org/project/mlfont/

Training Configuration

• Target Font: NotoSansSC-Regular.otf (30,759 Unicode codepoints)
• Training & Evaluation Corpus: wikimedia/wikipedia (20231101, languages: zh), 1,600 training articles and 400 uniformly sampled testing articles
• Slicing Architecture: Tiered Document-Frequency Slicing + Online Variational LDA & Deep Neural Network Topic Model (DNNTM) + Range-Compact Genetic Algorithm (mlfont preset: chinese)
• Total Slices: 102 slices (5,935 total CSS unicode-range intervals)
• Physical WOFF2 Bundle Size: 8.89 MB / 9107.9 KB (9,326,524 bytes) across 102 .woff2 files (companion CSS: 83.9 KB)
• Training Wall-Time: 87.6 seconds on CPU/GPU

Model Files Included

• model.safetensors: Learned topic-codepoint weight matrices (lambda_matrix, dnntm_lambda_matrix), codepoints, remaining_codepoints, frequency_array, and dna_assignment in SafeTensors format
• strategy.json: Range-Compact GA SlicingStrategy definition and hyperparameters
• olda_strategy.json: Online LDA + Greedy SlicingStrategy definition
• dnntm_strategy.json: DNNTM (Entropy Difference Loss) + Greedy SlicingStrategy definition
• test_samples.json: Exact held-out evaluation codepoint sets (400 test pages)
• config.json: Model and font bundle summary metadata (including per-slice WOFF2 byte sizes)
• font_sliced.css: Production-ready @font-face stylesheet with compact unicode-range descriptors
• woff2/*.woff2: 102 physical WOFF2 font slices generated via fontTools.subset
• nam/*.nam: 102 Google Fonts .nam codepoint subset files
• evaluation_metrics.json: Full benchmark metrics across all 7 network/session profiles

Benchmark Performance (400 Wikipedia Test Pages, Session Length L = 10)

Strategy	Slices	CSS Ranges	Missing (Eval / Font)	GIN-2g Mean / P95 (ms)	GIN-3g Mean / P95 (ms)	DSL Mean / P95 (ms)	Session Util (%)
single_slice	1	418	0 / 0	253.67 / 1629.56	63.47 / 407.72	12.68 / 81.45	5.55%
equal_bin_102	102	519	0 / 0	169.75 / 1165.36	45.26 / 310.69	8.21 / 56.33	7.63%
random_slice_102	102	30457	0 / 0	242.74 / 1930.38	64.71 / 514.64	11.73 / 93.32	5.55%
language_slice	10	6646	126 / 19659	86.54 / 554.31	21.82 / 139.59	4.31 / 27.61	16.05%
most_1000_500_equal_bin_100	102	2965	0 / 0	110.52 / 502.64	29.38 / 131.07	5.35 / 24.59	7.82%
google_fonts_ea_2017	102	6063	0 / 0	115.96 / 609.01	30.29 / 160.05	5.67 / 29.67	7.74%
online_lda_topics100_greedy	102	10177	0 / 0	60.16 / 314.64	15.93 / 83.23	2.92 / 15.27	16.90%
dnntm_topics100_greedy	102	9640	0 / 0	59.30 / 311.97	15.71 / 82.49	2.88 / 15.14	16.57%
range_compact_most1000_500_topics100 (This Model)	102	5935	0 / 0	66.67 / 356.73	17.67 / 94.33	3.23 / 17.32	14.64%

Usage

Install mlfont:

pip install mlfont huggingface_hub

Load the pretrained bundle (Range-Compact GA, Online LDA, and DNNTM) in Python:

import mlfont

bundle = mlfont.load_pretrained("wq2012/mlfont-notosanssc")
print("GA:", bundle.strategy.name, "ranges:", bundle.strategy.count_unicode_ranges())
print("LDA:", bundle.olda_strategy.name, "ranges:", bundle.olda_strategy.count_unicode_ranges())
print("DNNTM:", bundle.dnntm_strategy.name, "ranges:", bundle.dnntm_strategy.count_unicode_ranges())

Citation

@article{wang2017learning,
  title={Learning Better Font Slicing Strategies from Data},
  author={Wang, Quan and Mao, Yiran},
  journal={Technical Disclosure Commons, Defensive Publications Series},
  volume={906},
  year={2017}
}