mlfont-notosansjp: Range-Compact MLFont Slicing for Noto Sans JP (Japanese)

This is not an officially supported Google product.

Overview

Pretrained MLFont machine-learning-guided font slicing model and WOFF2 slice bundle for Noto Sans JP (Japanese) (NotoSansJP-Regular.otf, 16,663 Unicode codepoints partitioned into 62 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/

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 JP (Japanese) (NotoSansJP-Regular.otf) are much larger (4.67 MB / 4779.7 KB (4,894,460 bytes) in WOFF2) because they include 16,663 characters. In any large corpus of documents, there are strong co-occurrence patterns in character usage. By scanning webpages in ja, MLFont models which characters are most likely to appear together, informing a static subset slicing strategy that sorts all 16,663 glyphs into 62 slices (7,298 compact CSS unicode-range intervals).

Using this method, your entire web page renders in Noto Sans JP (Japanese) 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-notosansjp/font_sliced_cdn.css">

Or in your CSS / SCSS stylesheet:

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

2. Designate Font Family in CSS

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

body {
  font-family: 'Noto Sans JP Sliced', 'NotoSansJP-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-notosansjp/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="notosansjp"
        async></script>

Or programmatically in JavaScript:

MLFont.load('notosansjp');

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

Included in this bundle (mlfont.js, an open-source reproduction and extension of Google Fonts' //fonts.gstatic.com/ea/timing/v1/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 JP (Japanese) (full unsliced WOFF2: 3,260.2 KB), range_compact_most500_500_topics60 downloads an average of only 10.33 / 62 slices (1,062.9 KB vs. 3,279.1 KB unsliced, saving 67.6% bandwidth). On Cable-5Mbps (5 Mbps, 28 ms RTT), this achieves a 2.94x cold-cache load speedup (1,851 ms vs. 5,437 ms for single_slice), with First Text Paint at 689 ms and Warm-Session Load at 568 ms (and 2.94x speedup on GIN-3g: 23,117 ms vs. 67,965 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	3,279.1 KB	0.0%	5,437 ms	5,437 ms	0 ms	67,965 ms	1.00x / 1.00x
equal_bin_62	37.00 / 62	2,873.3 KB	12.4%	859 ms	4,924 ms	989 ms	61,789 ms	1.10x / 1.10x
most_500_500_equal_bin_60	14.00 / 62	1,270.6 KB	61.2%	678 ms	2,207 ms	1,727 ms	27,636 ms	2.46x / 2.46x
language_slice	2.00 / 10	1,949.9 KB	40.5%	3,261 ms	3,261 ms	0 ms	40,745 ms	1.67x / 1.67x
online_lda_topics60_greedy	9.33 / 62	930.8 KB	71.6%	696 ms	1,628 ms	521 ms	20,281 ms	3.34x / 3.35x
dnntm_topics60_greedy	7.67 / 62	805.1 KB	75.5%	695 ms	1,412 ms	644 ms	17,573 ms	3.85x / 3.87x
range_compact_most500_500_topics60 (This Model)	10.33 / 62	1,062.9 KB	67.6%	689 ms	1,851 ms	568 ms	23,117 ms	2.94x / 2.94x

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

Verified against physical .woff2 files in Headless Google 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)
jp_web_dev_docs	single_slice	1	Yes	3,279.1 KB	165.7 ms	5,594 ms	1.00x	1.00x (0.0%)
jp_web_dev_docs	equal_bin_62	34	Yes	2,642.4 KB	91.7 ms	4,617 ms	1.81x	1.21x (19.4%)
jp_web_dev_docs	range_compact_most500_500_topics60 (This Model)	7	Yes	773.6 KB	79.1 ms	1,431 ms	2.09x	3.91x (76.4%)
jp_sports_baseball	single_slice	1	Yes	3,279.1 KB	164.7 ms	5,593 ms	1.00x	1.00x (0.0%)
jp_sports_baseball	equal_bin_62	35	Yes	2,718.8 KB	93.5 ms	4,744 ms	1.76x	1.18x (17.1%)
jp_sports_baseball	range_compact_most500_500_topics60 (This Model)	10	Yes	1,054.6 KB	63.6 ms	1,875 ms	2.59x	2.98x (67.8%)
jp_literature_essay	single_slice	1	Yes	3,279.1 KB	250.8 ms	5,679 ms	1.00x	1.00x (0.0%)
jp_literature_essay	equal_bin_62	42	Yes	3,258.6 KB	264.9 ms	5,828 ms	0.95x	0.97x (0.6%)
jp_literature_essay	range_compact_most500_500_topics60 (This Model)	14	Yes	1,360.4 KB	198.7 ms	2,540 ms	1.26x	2.24x (58.5%)

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

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

Training Configuration

• Target Font: NotoSansJP-Regular.otf (16,663 Unicode codepoints)
• Training & Evaluation Corpus: wikimedia/wikipedia (20231101, languages: ja), 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: japanese)
• Total Slices: 62 slices (7,298 total CSS unicode-range intervals)
• Physical WOFF2 Bundle Size: 4.67 MB / 4779.7 KB (4,894,460 bytes) across 62 .woff2 files (companion CSS: 85.3 KB)
• Training Wall-Time: 46.0 seconds on CPU/GPU

Model Files Included

• font_sliced_cdn.css: Zero-hosting CDN @font-face stylesheet (font-family: 'Noto Sans JP Sliced') streaming .woff2 slices directly from Hugging Face CDN
• font_sliced.css: Self-hosted @font-face stylesheet (font-family: 'NotoSansJP-Regular Sliced') with compact unicode-range descriptors
• mlfont.js: Browser JS loader (MLFont.load) & HTML5 Performance API slice latency telemetry library (MLFont.getMetrics / window.GoogleFonts)
• woff2/*.woff2: 62 physical WOFF2 font slices generated via fontTools.subset
• nam/*.nam: 62 Google Fonts .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	4830	0 / 0	94.75 / 701.82	23.72 / 175.72	4.73 / 35.06	7.62%
equal_bin_62	62	4876	0 / 0	95.38 / 698.56	25.58 / 187.34	4.60 / 33.66	8.56%
random_slice_62	62	16450	0 / 0	114.88 / 893.64	30.81 / 239.66	5.53 / 43.06	7.62%
language_slice	10	5617	43 / 7632	45.20 / 305.76	11.42 / 77.00	2.25 / 15.22	15.84%
most_500_500_equal_bin_60	62	5975	0 / 0	76.50 / 387.35	20.50 / 103.64	3.69 / 18.69	8.92%
online_lda_topics60_greedy	62	8888	0 / 0	34.87 / 201.00	9.31 / 53.61	1.68 / 9.71	21.31%
dnntm_topics60_greedy	62	8710	0 / 0	34.80 / 188.72	9.30 / 50.47	1.68 / 9.12	20.88%
range_compact_most500_500_topics60 (This Model)	62	7298	0 / 0	42.53 / 223.47	11.37 / 59.70	2.05 / 10.79	16.91%

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-notosansjp")
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}
}