mlfont-notosans-lgc: Range-Compact MLFont Slicing for Noto Sans (Pan-European LGC)

Overview

Pretrained MLFont machine-learning-guided font slicing model and WOFF2 slice bundle for Noto Sans (Pan-European LGC) (NotoSans-Regular.ttf, 2,297 Unicode codepoints partitioned into 20 range-compact slices).

• Paper: MLFont: Machine Learning-Driven Font Subsetting and Range-Compact Unicode-Range Slicing for Low-Latency Web Typography (PDF) — Quan Wang and Yiran Mao (arXiv:2610.24937)
• Background & References: Open-source reproduction of Google's machine learning font slicing technology (Official Post, Defensive Publication)
• Open-Source Repository: https://github.com/wq2012/mlfont
• PyPI Package: https://pypi.org/project/mlfont/

Using This Sliced Font in Your Website (Frontend Developer Guide)

While basic Latin webfonts are typically 20–100 KB, Pan-Unicode and East Asian fonts like Noto Sans (Pan-European LGC) (NotoSans-Regular.ttf) are much larger (434.1 KB (444,484 bytes) in WOFF2) because they include 2,297 characters. In any large corpus of documents, there are strong co-occurrence patterns in character usage. By scanning webpages in en, de, fr, pl, ru, el, MLFont models which characters are most likely to appear together, informing a static subset slicing strategy that sorts all 2,297 glyphs into 20 slices (384 compact CSS unicode-range intervals).

Using this method, your entire web page renders in Noto Sans (Pan-European LGC) while the browser only downloads the small subset of .woff2 font slices required for the characters present on that page—which is dramatically faster than loading the whole font. Cross-page and cross-site browser caching further increases the latency benefits over time. (Note: The slicing optimization works automatically in all modern browsers that support CSS unicode-range.)

1. Link to Sliced Font CSS (HTML <link> or CSS @import)

Add a <link> element to the <head> section of your HTML:

<link rel="stylesheet"
      href="https://cdn.jsdelivr.net/gh/wq2012/mlfont@main/pretrained_models/mlfont-notosans-lgc/font_sliced_cdn.css">

Or in your CSS / SCSS stylesheet:

@import url('https://cdn.jsdelivr.net/gh/wq2012/mlfont@main/pretrained_models/mlfont-notosans-lgc/font_sliced_cdn.css');

2. Designate Font Family in CSS

Style your text with the 'Noto Sans Sliced' family in your CSS:

body {
  font-family: 'Noto Sans Sliced', 'NotoSans-Regular Sliced', Roboto, 'Noto Sans', sans-serif;
}

(Note: font_sliced_cdn.css streams the .woff2 slices on-demand directly from Hugging Face's global CDN https://huggingface.co/wq2012/mlfont-notosans-lgc/resolve/main/woff2/... with font-display: swap and CORS Access-Control-Allow-Origin: *, so you do not need to host the .woff2 files yourself. To self-host on your own server/CDN, download this repository and link font_sliced.css alongside the woff2/ folder.)

3. Alternative: One-Line Drop-In JavaScript Loader (mlfont.js)

You can also load the sliced stylesheet directly from Hugging Face CDN using a single <script> tag with data-mlfont:

<script src="https://cdn.jsdelivr.net/gh/wq2012/mlfont@main/web/mlfont.js"
        data-mlfont="notosans-lgc"
        async></script>

Or programmatically in JavaScript:

MLFont.load('notosans-lgc');

4. Latency Metrics & Browser Telemetry (Recommended — mlfont.js)

Included in this bundle (mlfont.js) is a small, zero-dependency JS library that accesses the HTML5 Performance API to measure webfont slice performance. No user-specific data is collected. The JS library collects:

1. Download time metrics of individual font slices (slice index, TTFB, download duration, transfer bytes, cache hit status)
2. Download time of the sliced @font-face CSS file
3. Page onload and First Contentful Paint (FCP) time

To include the library, add the following code to the bottom of your <head> section:

<script src="https://cdn.jsdelivr.net/gh/wq2012/mlfont@main/web/mlfont.js" async></script>

You can inspect downloaded slice metrics in your browser DevTools console via MLFont.printReport() or MLFont.getMetrics(), or forward anonymized telemetry beacons to your own analytics endpoint via data-endpoint="https://your-analytics.example.com/mlfont-timing".

5. Simulating Real-Website Font Loading Latency & Headless Chrome Verification

Across 3 real-world sample websites for Noto Sans (Pan-European LGC) (full unsliced WOFF2: 157.7 KB), range_compact_topics20 downloads an average of only 4.00 / 20 slices (87.6 KB vs. 158.4 KB unsliced, saving 44.7% bandwidth). On Cable-5Mbps (5 Mbps, 28 ms RTT), this achieves a 1.57x cold-cache load speedup (201 ms vs. 316 ms for single_slice), with First Text Paint at 201 ms and Warm-Session Load at 0 ms (and 1.56x speedup on GIN-3g: 2,596 ms vs. 4,043 ms):

Strategy	Slices Dl / Total	Transfer (KB)	Bytes Saved	Cable-5Mbps 1st Paint (ms)	Cable-5Mbps Cold Load (ms)	Cable-5Mbps Warm Load (ms)	GIN-3g Cold Load (ms)	Cold Speedup (Cable / 3G)
single_slice	1.00 / 1	158.4 KB	0.0%	316 ms	316 ms	0 ms	4,043 ms	1.00x / 1.00x
equal_bin_20	4.00 / 20	87.8 KB	44.6%	201 ms	201 ms	32 ms	2,599 ms	1.57x / 1.56x
language_slice	3.00 / 8	138.8 KB	12.4%	284 ms	284 ms	49 ms	3,642 ms	1.11x / 1.11x
online_lda_topics20_greedy	3.67 / 20	82.0 KB	48.2%	192 ms	192 ms	0 ms	2,480 ms	1.64x / 1.63x
dnntm_topics20_greedy	3.33 / 20	74.7 KB	52.8%	180 ms	180 ms	32 ms	2,333 ms	1.75x / 1.73x
range_compact_topics20 (This Model)	4.00 / 20	87.6 KB	44.7%	201 ms	201 ms	0 ms	2,596 ms	1.57x / 1.56x

Real Headless Chrome (v155.0.8059.79) Verification (Cable-5Mbps)

Verified against physical .woff2 files in Headless Chrome (--headless=new --dump-dom) via the W3C document.fonts (FontFaceSet) API (9/9 = 100.0% exact slice-set match between Chromium Blink's unicode-range engine and WebsiteFontSimulator).

Website Page ID	Strategy	Chrome Loaded Slices	Sim Exact Match	Transfer (KB)	Chrome Local Decode (ms)	E2E Font Ready (ms)	Decode Speedup	E2E Speedup (Bytes Saved)
lgc_multilingual_eu_news	single_slice	1	Yes	158.4 KB	28.6 ms	344 ms	1.00x	1.00x (0.0%)
lgc_multilingual_eu_news	equal_bin_20	6	Yes	131.2 KB	27.2 ms	298 ms	1.05x	1.15x (17.2%)
lgc_multilingual_eu_news	range_compact_topics20 (This Model)	5	Yes	110.5 KB	24.5 ms	262 ms	1.17x	1.32x (30.2%)
lgc_slavic_hellenic_encyclopedia	single_slice	1	Yes	158.4 KB	30.9 ms	346 ms	1.00x	1.00x (0.0%)
lgc_slavic_hellenic_encyclopedia	equal_bin_20	5	Yes	109.4 KB	29.3 ms	265 ms	1.05x	1.31x (30.9%)
lgc_slavic_hellenic_encyclopedia	range_compact_topics20 (This Model)	5	Yes	110.5 KB	27.1 ms	264 ms	1.14x	1.31x (30.2%)
lgc_scientific_typography	single_slice	1	Yes	158.4 KB	22.5 ms	338 ms	1.00x	1.00x (0.0%)
lgc_scientific_typography	equal_bin_20	1	Yes	22.7 KB	17.9 ms	111 ms	1.26x	3.04x (85.7%)
lgc_scientific_typography	range_compact_topics20 (This Model)	2	Yes	41.7 KB	18.7 ms	143 ms	1.20x	2.36x (73.7%)

You can test your own HTML files, live URLs, or text snippets against mlfont-notosans-lgc before deploying using the mlfont-simulate CLI, the Python WebsiteFontSimulator API, or the interactive side-by-side Website Font Loading Simulator in web/demo.html:

# Simulate a local HTML file with mlfont-notosans-lgc across all 7 network profiles:
mlfont-simulate --font_key notosans_lgc --html_file path/to/index.html --network_profile all

# Simulate a live URL and verify in real Headless Chrome:
mlfont-simulate --font_key notosans_lgc --url https://example.com --network_profile Cable-5Mbps --headless_chrome

Training Configuration

• Target Font: NotoSans-Regular.ttf (2,297 Unicode codepoints)
• Training & Evaluation Corpus: wikimedia/wikipedia (20231101, languages: en, de, fr, pl, ru, el), 1,500 training articles and 300 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: lgc)
• Total Slices: 20 slices (384 total CSS unicode-range intervals)
• Physical WOFF2 Bundle Size: 434.1 KB (444,484 bytes) across 20 .woff2 files (companion CSS: 7.9 KB). Note: physical .woff2 slicing replicates shared OpenType tables (cmap, hmtx, loca, GSUB, GPOS) into each standalone subset file (increasing static CDN disk footprint by +46.6% to +175.3% vs. a monolithic WOFF2), whereas individual webpages only fetch the small subset of triggered slices (saving 65.4%–84.9% per-page bandwidth on CJK and 44.7% on LGC).
• Training Wall-Time: 24.0 seconds on CPU/GPU

Model Files Included

• font_sliced_cdn.css: Zero-hosting CDN @font-face stylesheet (font-family: 'Noto Sans Sliced') streaming .woff2 slices directly from Hugging Face CDN
• font_sliced.css: Self-hosted @font-face stylesheet (font-family: 'NotoSans-Regular Sliced', url(woff2/NotoSans-Regular_slice_XXX.woff2)) with compact unicode-range descriptors
• mlfont.js: Browser JS loader (MLFont.load) & HTML5 Performance API slice latency telemetry library (MLFont.getMetrics / MLFont.printReport)
• woff2/NotoSans-Regular_slice_XXX.woff2: 20 physical WOFF2 font slices generated via fontTools.subset (plus woff2/NotoSans-Regular_sliced.css)
• nam/slice_XXX.nam: 20 .nam codepoint subset files
• 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 (300 test pages)
• config.json: Model and font bundle summary metadata (including per-slice WOFF2 byte sizes)
• evaluation_metrics.json: Full benchmark metrics across all 7 network/session profiles

Benchmark Performance (300 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	123	0 / 0	60.26 / 566.45	15.22 / 143.10	3.00 / 28.17	8.66%
equal_bin_20	20	140	0 / 0	33.37 / 153.43	9.44 / 43.41	1.56 / 7.17	23.17%
random_slice_20	20	2180	0 / 0	95.82 / 906.75	27.11 / 256.54	4.48 / 42.35	8.66%
language_slice	8	92	6 / 300	40.17 / 258.54	10.62 / 66.28	1.95 / 12.76	15.08%
online_lda_topics20_greedy	20	537	0 / 0	25.15 / 144.76	7.11 / 40.95	1.17 / 6.76	31.26%
dnntm_topics20_greedy	20	549	0 / 0	23.29 / 112.70	6.59 / 31.88	1.09 / 5.26	32.56%
range_compact_topics20 (This Model)	20	384	0 / 0	25.05 / 142.01	7.06 / 40.06	1.17 / 6.64	30.31%

Python API Usage (ML Researchers & Custom Font Slicing)

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-notosans-lgc")
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{wang2026mlfont,
  title={MLFont: Machine Learning-Driven Font Subsetting and Range-Compact Unicode-Range Slicing for Low-Latency Web Typography},
  author={Wang, Quan and Mao, Yiran},
  journal={arXiv preprint arXiv:2610.24937},
  eprint={2610.24937},
  archivePrefix={arXiv},
  year={2026},
  url={https://arxiv.org/abs/2610.24937}
}

@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},
  url={https://www.tdcommons.org/dpubs_series/906/}
}