gnumanth commited on
Commit
785c0c2
·
verified ·
1 Parent(s): c549f81

Upload ScribeFont FP32 safetensors + Int8 quantized ONNX + empirical report

Browse files
README.md CHANGED
@@ -39,7 +39,7 @@ Following the 3-tier scientific module playbook, `scribefont` separates **2D Con
39
  | :--- | :--- | :---: | :---: | :--- |
40
  | **Tier 1A: Neural Style-Conditioned (FP32)** | [`gnumanth/scribefont`](https://huggingface.co/gnumanth/scribefont) (`model.safetensors`) | `11.4 MB` | `2.94M` (`Conv2d` + `4-Layer Transformer`) | Few-shot unseen glyph synthesis from noisy camera scans, writer style extraction |
41
  | **Tier 1B: Neural Quantized (Int8 ONNX)** | [`onnx/model_quantized.onnx`](https://huggingface.co/gnumanth/scribefont/tree/main/onnx) (`onnxruntime-web` WASM SIMD / CPU) | `3.15 MB` (`3.62x` smaller) | `2.94M` (`QUInt8` weights) | **100% Client-Side Browser WASM SIMD / WebGPU** handwriting-to-`.ttf` studio |
42
- | **Tier 2: Zero-Dep Fast Engine** | `npm i scribefont` / `pip install scribefont` | `0.0 MB` (`~18 KB` code) | `0` (Deterministic Bézier + `.ttf` packer) | Ultra-low-latency (`0.0317 ms/glyph`) parametric glyph synthesis & binary `.ttf` compilation |
43
 
44
  ---
45
 
@@ -49,14 +49,14 @@ Evaluates synthesizing unseen target characters (`30` valid unseen glyphs + `30`
49
 
50
  | Metric ($N=60$, $B=1,000$ Bootstrap 95% CI) | Tier 1A: PyTorch FP32 Neural | Tier 1B: ONNX Int8 Quantized (WASM/CPU) | Tier 2: Zero-Dep Geometric Engine |
51
  | :--- | :---: | :---: | :---: |
52
- | **Overall Accuracy (95% Bootstrap CI)** | **100.0%** `[100.0, 100.0]` | **100.0%** `[100.0, 100.0]` | 73.33% `[61.67, 83.33]` |
53
- | **Unseen-Glyph Style Fidelity (`MAE < 12 UPM`)** | **100.0%** `[100.0, 100.0]` | **100.0%** `[100.0, 100.0]` | 46.67% `[26.67, 63.33]` |
54
- | **Bézier Contour Coordinate MAE (`UPM`, lower is better)** | **2.14 UPM** | **2.15 UPM** | 12.71 UPM |
55
- | **Advance-Width Prediction MAE (`UPM`)** | **10.61 UPM** | **10.63 UPM** | 39.67 UPM |
56
  | **12-Writer Calligrapher Identification Accuracy** | **100.0%** | **100.0%** | 16.67% |
57
  | **Blank/Smudge Cell Abstention (`p_missing > 0.5`)** | **100.0%** | **100.0%** | **100.0%** |
58
- | **Abstention `p_missing` AUROC / 10-Bin ECE** | **1.0** / `0.00759` | **1.0** / `0.00749` | **1.0** / `0.0006` |
59
- | **Latency `p50` / `p95` / `p99` (ms/glyph)** | `7.6123` / `11.4627` / `12.017 ms` | `1.2025` / `1.9507` / `2.046 ms` | **0.0307** / `0.033` / `0.0397 ms` |
60
 
61
  ---
62
 
@@ -67,10 +67,10 @@ Evaluates full-alphabet (`A-Z`, `a-z`, `0-9`) glyph contour synthesis, metric in
67
  | Metric ($N=62$ alphanumeric glyphs) | Tier 1A: PyTorch FP32 Neural | Tier 1B: ONNX Int8 Quantized | Tier 2: Zero-Dep Fast Engine |
68
  | :--- | :---: | :---: | :---: |
69
  | **Glyph Fidelity Accuracy (95% Bootstrap CI)** | **100.0%** `[100.0, 100.0]` | **100.0%** `[100.0, 100.0]` | **100.0%** `[100.0, 100.0]` |
70
- | **Bézier Contour MAE (`UPM`, out of `1000 UPM`)** | `1.11 UPM` | `1.12 UPM` | **0.0 UPM** |
71
  | **OpenType Metric Invariant (`adv >= w + lsb`)** | **100.0%** | **100.0%** | **100.0%** |
72
  | **Compiled Binary `.ttf` Checksum & SFNT Validity** | **100.0%** (`7292 B`) | **100.0%** (`7292 B`) | **100.0%** (`7292 B`) |
73
- | **Per-Glyph Latency `p50` / `p95` (ms)** | `4.3805` / `6.3712 ms` | `0.9642` / `1.5815 ms` | **0.0317** / `0.0377 ms` |
74
 
75
  ---
76
 
@@ -80,5 +80,5 @@ Evaluates full-alphabet (`A-Z`, `a-z`, `0-9`) glyph contour synthesis, metric in
80
  | :--- | :---: | :---: | :---: |
81
  | **Adversarial Glyph & Abstention Accuracy (95% CI)** | **100.0%** `[100.0, 100.0]` | **100.0%** `[100.0, 100.0]` | **100.0%** `[100.0, 100.0]` |
82
  | **Calibrated `p_missing` AUROC** | **1.0** | **1.0** | **1.0** |
83
- | **Abstention 10-Bin ECE / Brier Score** | `0.00104` / `0.0` | `0.00106` / `0.0` | `0.0006` / `0.0` |
84
 
 
39
  | :--- | :--- | :---: | :---: | :--- |
40
  | **Tier 1A: Neural Style-Conditioned (FP32)** | [`gnumanth/scribefont`](https://huggingface.co/gnumanth/scribefont) (`model.safetensors`) | `11.4 MB` | `2.94M` (`Conv2d` + `4-Layer Transformer`) | Few-shot unseen glyph synthesis from noisy camera scans, writer style extraction |
41
  | **Tier 1B: Neural Quantized (Int8 ONNX)** | [`onnx/model_quantized.onnx`](https://huggingface.co/gnumanth/scribefont/tree/main/onnx) (`onnxruntime-web` WASM SIMD / CPU) | `3.15 MB` (`3.62x` smaller) | `2.94M` (`QUInt8` weights) | **100% Client-Side Browser WASM SIMD / WebGPU** handwriting-to-`.ttf` studio |
42
+ | **Tier 2: Zero-Dep Fast Engine** | `npm i scribefont` / `pip install scribefont` | `0.0 MB` (`~18 KB` code) | `0` (Deterministic Bézier + `.ttf` packer) | Ultra-low-latency (`0.0747 ms/glyph`) parametric glyph synthesis & binary `.ttf` compilation |
43
 
44
  ---
45
 
 
49
 
50
  | Metric ($N=60$, $B=1,000$ Bootstrap 95% CI) | Tier 1A: PyTorch FP32 Neural | Tier 1B: ONNX Int8 Quantized (WASM/CPU) | Tier 2: Zero-Dep Geometric Engine |
51
  | :--- | :---: | :---: | :---: |
52
+ | **Overall Accuracy (95% Bootstrap CI)** | **100.0%** `[100.0, 100.0]` | **100.0%** `[100.0, 100.0]` | 80.0% `[70.0, 90.0]` |
53
+ | **Unseen-Glyph Style Fidelity (`MAE < 12 UPM`)** | **100.0%** `[100.0, 100.0]` | **100.0%** `[100.0, 100.0]` | 60.0% `[40.0, 76.67]` |
54
+ | **Bézier Contour Coordinate MAE (`UPM`, lower is better)** | **2.38 UPM** | **2.39 UPM** | 11.13 UPM |
55
+ | **Advance-Width Prediction MAE (`UPM`)** | **13.93 UPM** | **14.02 UPM** | 35.5 UPM |
56
  | **12-Writer Calligrapher Identification Accuracy** | **100.0%** | **100.0%** | 16.67% |
57
  | **Blank/Smudge Cell Abstention (`p_missing > 0.5`)** | **100.0%** | **100.0%** | **100.0%** |
58
+ | **Abstention `p_missing` AUROC / 10-Bin ECE** | **1.0** / `0.00276` | **1.0** / `0.00282` | **1.0** / `0.0006` |
59
+ | **Latency `p50` / `p95` / `p99` (ms/glyph)** | `6.1797` / `8.3667` / `9.2688 ms` | `1.2708` / `1.5253` / `1.5695 ms` | **0.0539** / `0.0632` / `0.0716 ms` |
60
 
61
  ---
62
 
 
67
  | Metric ($N=62$ alphanumeric glyphs) | Tier 1A: PyTorch FP32 Neural | Tier 1B: ONNX Int8 Quantized | Tier 2: Zero-Dep Fast Engine |
68
  | :--- | :---: | :---: | :---: |
69
  | **Glyph Fidelity Accuracy (95% Bootstrap CI)** | **100.0%** `[100.0, 100.0]` | **100.0%** `[100.0, 100.0]` | **100.0%** `[100.0, 100.0]` |
70
+ | **Bézier Contour MAE (`UPM`, out of `1000 UPM`)** | `0.98 UPM` | `0.98 UPM` | **0.0 UPM** |
71
  | **OpenType Metric Invariant (`adv >= w + lsb`)** | **100.0%** | **100.0%** | **100.0%** |
72
  | **Compiled Binary `.ttf` Checksum & SFNT Validity** | **100.0%** (`7292 B`) | **100.0%** (`7292 B`) | **100.0%** (`7292 B`) |
73
+ | **Per-Glyph Latency `p50` / `p95` (ms)** | `6.035` / `9.1311 ms` | `1.3858` / `1.9749 ms` | **0.0747** / `0.083 ms` |
74
 
75
  ---
76
 
 
80
  | :--- | :---: | :---: | :---: |
81
  | **Adversarial Glyph & Abstention Accuracy (95% CI)** | **100.0%** `[100.0, 100.0]` | **100.0%** `[100.0, 100.0]` | **100.0%** `[100.0, 100.0]` |
82
  | **Calibrated `p_missing` AUROC** | **1.0** | **1.0** | **1.0** |
83
+ | **Abstention 10-Bin ECE / Brier Score** | `0.00077` / `0.0` | `0.00077` / `0.0` | `0.0006` / `0.0` |
84
 
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e26a83d8c3a963bdfdf7ce269f93c6e8772713867024123c13449bdd7df75cc7
3
  size 11776632
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0a0f4b181cc6a5d940e9df2c2f676e7fe18e12f770055156a67f7a3721ca1e15
3
  size 11776632
onnx/model.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ea69cd4111fe030fbc103ac5271b9df0e10fa3245e681c6da592c0b35a166ebe
3
  size 11954706
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d259cec17530d16f05b3598d8e788e28c54239ea6bbde09e7ad304a89ae08b12
3
  size 11954706
onnx/model_quantized.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:c2196fd9d45ae12859adc941cc377e8f8a0a598eb4ae5e87413d9930eb2a16bf
3
- size 3307675
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:66d970169c7801fdc26ccf956cd93db6f70e979761a714373330afbd3b34c190
3
+ size 3307680
training_summary.json CHANGED
@@ -2,11 +2,11 @@
2
  "model": "ScribeFontForGeneration",
3
  "parameters": 2941340,
4
  "device": "mps",
5
- "train_seconds": 32.38,
6
  "best_val_metrics": {
7
  "missing_acc": 1.0,
8
  "writer_acc": 1.0,
9
- "contour_mae_upm": 1.41,
10
- "advance_mae_upm": 2.83
11
  }
12
  }
 
2
  "model": "ScribeFontForGeneration",
3
  "parameters": 2941340,
4
  "device": "mps",
5
+ "train_seconds": 62.14,
6
  "best_val_metrics": {
7
  "missing_acc": 1.0,
8
  "writer_acc": 1.0,
9
+ "contour_mae_upm": 1.25,
10
+ "advance_mae_upm": 2.77
11
  }
12
  }