mlfont-notosanstc: Range-Compact MLFont Slicing for Noto Sans TC (Traditional Chinese)

Overview

Pretrained MLFont machine-learning-guided font slicing model and WOFF2 slice bundle for Noto Sans TC (Traditional Chinese) (NotoSansTC-Regular.otf, 20,637 Unicode codepoints partitioned into 82 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)
• 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 TC (Traditional Chinese) (NotoSansTC-Regular.otf) are much larger (5.94 MB / 6085.5 KB (6,231,532 bytes) in WOFF2) because they include 20,637 characters. In any large corpus of documents, there are strong co-occurrence patterns in character usage. By scanning webpages in zh, MLFont models which characters are most likely to appear together, informing a static subset slicing strategy that sorts all 20,637 glyphs into 82 slices (8,743 compact CSS unicode-range intervals).

Using this method, your entire web page renders in Noto Sans TC (Traditional Chinese) 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-notosanstc/font_sliced_cdn.css">

Or in your CSS / SCSS stylesheet:

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

2. Designate Font Family in CSS

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

body {
  font-family: 'Noto Sans TC Sliced', 'NotoSansTC-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-notosanstc/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="notosanstc"
        async></script>

Or programmatically in JavaScript:

MLFont.load('notosanstc');

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 TC (Traditional Chinese) (full unsliced WOFF2: 4,047.4 KB), range_compact_most600_600_topics80 downloads an average of only 16.00 / 82 slices (1,406.8 KB vs. 4,067.0 KB unsliced, saving 65.4% bandwidth). On Cable-5Mbps (5 Mbps, 28 ms RTT), this achieves a 2.76x cold-cache load speedup (2,436 ms vs. 6,728 ms for single_slice), with First Text Paint at 665 ms and Warm-Session Load at 737 ms (and 2.76x speedup on GIN-3g: 30,431 ms vs. 84,100 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	4,067.0 KB	0.0%	6,728 ms	6,728 ms	0 ms	84,100 ms	1.00x / 1.00x
equal_bin_82	52.00 / 82	3,881.7 KB	4.6%	834 ms	6,662 ms	589 ms	83,644 ms	1.01x / 1.01x
most_600_600_equal_bin_80	22.67 / 82	1,833.9 KB	54.9%	648 ms	3,170 ms	1,978 ms	39,707 ms	2.12x / 2.12x
language_slice	3.00 / 10	2,287.2 KB	43.8%	3,814 ms	3,814 ms	0 ms	47,652 ms	1.76x / 1.76x
online_lda_topics80_greedy	16.33 / 82	1,394.0 KB	65.7%	681 ms	2,430 ms	616 ms	30,307 ms	2.77x / 2.77x
dnntm_topics80_greedy	16.00 / 82	1,369.1 KB	66.3%	678 ms	2,389 ms	616 ms	29,796 ms	2.82x / 2.82x
range_compact_most600_600_topics80 (This Model)	16.00 / 82	1,406.8 KB	65.4%	665 ms	2,436 ms	737 ms	30,431 ms	2.76x / 2.76x

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% 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)
tc_taipei_tech_portal	single_slice	1	Yes	4,067.0 KB	172.4 ms	6,892 ms	1.00x	1.00x (0.0%)
tc_taipei_tech_portal	equal_bin_82	51	Yes	3,807.7 KB	104.3 ms	6,623 ms	1.65x	1.04x (6.4%)
tc_taipei_tech_portal	range_compact_most600_600_topics80 (This Model)	12	Yes	1,103.5 KB	54.2 ms	1,946 ms	3.18x	3.54x (72.9%)
tc_culture_literature	single_slice	1	Yes	4,067.0 KB	165.4 ms	6,885 ms	1.00x	1.00x (0.0%)
tc_culture_literature	equal_bin_82	55	Yes	4,104.1 KB	105.2 ms	7,137 ms	1.57x	0.96x (-0.9%)
tc_culture_literature	range_compact_most600_600_topics80 (This Model)	16	Yes	1,389.4 KB	54.7 ms	2,443 ms	3.02x	2.82x (65.8%)
tc_gastronomy_guide	single_slice	1	Yes	4,067.0 KB	175.8 ms	6,895 ms	1.00x	1.00x (0.0%)
tc_gastronomy_guide	equal_bin_82	50	Yes	3,733.4 KB	125.8 ms	6,523 ms	1.40x	1.06x (8.2%)
tc_gastronomy_guide	range_compact_most600_600_topics80 (This Model)	20	Yes	1,727.6 KB	62.3 ms	3,033 ms	2.82x	2.27x (57.5%)

You can test your own HTML files, live URLs, or text snippets against mlfont-notosanstc 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-notosanstc across all 7 network profiles:
mlfont-simulate --font_key notosanstc --html_file path/to/index.html --network_profile all

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

Training Configuration

• Target Font: NotoSansTC-Regular.otf (20,637 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: traditional_chinese)
• Total Slices: 82 slices (8,743 total CSS unicode-range intervals)
• Physical WOFF2 Bundle Size: 5.94 MB / 6085.5 KB (6,231,532 bytes) across 82 .woff2 files (companion CSS: 104.1 KB)
• Training Wall-Time: 25.0 seconds on CPU/GPU

Model Files Included

• font_sliced_cdn.css: Zero-hosting CDN @font-face stylesheet (font-family: 'Noto Sans TC Sliced') streaming .woff2 slices directly from Hugging Face CDN
• font_sliced.css: Self-hosted @font-face stylesheet (font-family: 'NotoSansTC-Regular Sliced') with compact unicode-range descriptors
• mlfont.js: Browser JS loader (MLFont.load) & HTML5 Performance API slice latency telemetry library (MLFont.getMetrics / MLFont.printReport)
• woff2/*.woff2: 82 physical WOFF2 font slices generated via fontTools.subset
• nam/*.nam: 82 .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 (400 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 (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	5036	0 / 0	125.58 / 704.12	31.43 / 176.24	6.28 / 35.18	7.88%
equal_bin_82	82	5098	0 / 0	115.29 / 689.78	31.03 / 185.64	5.54 / 33.17	9.67%
random_slice_82	82	20436	0 / 0	144.57 / 922.42	38.91 / 248.25	6.95 / 44.36	7.88%
language_slice	10	5767	126 / 11699	52.25 / 282.95	13.23 / 71.40	2.60 / 14.08	18.89%
most_600_600_equal_bin_80	82	6672	0 / 0	92.66 / 446.08	24.90 / 119.91	4.46 / 21.47	9.83%
online_lda_topics80_greedy	82	11867	0 / 0	50.11 / 250.61	13.43 / 67.58	2.42 / 12.04	19.59%
dnntm_topics80_greedy	82	11410	0 / 0	51.65 / 226.65	13.84 / 60.72	2.49 / 10.93	19.30%
range_compact_most600_600_topics80 (This Model)	82	8743	0 / 0	56.19 / 265.70	15.07 / 71.62	2.71 / 12.79	17.37%

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-notosanstc")
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}
}