r2s-v10-img192-5247da7-bcf071a7-360aa736
completed —
Hypothesis
With input renders 144 -> 192 px (image grid 18 x 18 -> 24 x 24), and everything else as in v9, the render-to-program model transcribes the 339 validation renders with lower mean 72 px pixel error than v9 trained on the same machine and seed. Pass requires the paired 95% bootstrap interval on the per-icon reduction against that v9 rerun to exclude zero. Its greedy latency on the RTX 4080 is recorded; a pass that is too slow for the browser demo is a teacher for distillation, not the demo model.
Visual output
Written result
Hypothesis. With input renders 144 -> 192 px (image grid 18 x 18 -> 24 x 24), and everything else as in v9, the render-to-program model transcribes the 339 validation renders with lower mean 72 px pixel error than v9 trained on the same machine and seed. Pass requires the paired 95% bootstrap interval on the per-icon reduction against that v9 rerun to exclude zero. Its greedy latency on the RTX 4080 is recorded; a pass that is too slow for the browser demo is a teacher for distillation, not the demo model.
Result
| measure | value |
|---|---|
| icons evaluated | 339 (primary/validation) |
| model pixel error, mean [95% CI] | 0.0778 [0.0723, 0.0833] |
| model pixel error, median | 0.0739 |
| exact program rate | 0.021 |
| rendered rate | 1.000 |
| blank canvas pixel error | 0.1720 |
| nearest training icon pixel error | 0.0897 [0.0848, 0.0947] |
| error reduction vs nearest icon | 0.0119 [0.0077, 0.0161] |
| icons where model beats nearest icon | 217 |
| CLIP top-1, model greedy (32 Gate N icons) | 0.656 |
| CLIP top-1, nearest training icon | 0.344 |
| CLIP top-1, OmniSVG 4B zero-shot (Gate N) | 0.609 |
Resources
| measure | value |
|---|---|
| device | NVIDIA H100 |
| parameters | 9,090,338 |
| train seconds | 3431 |
| peak VRAM GiB | 4.671260833740234 |
| latency ms/icon, batch 1, median | 3652.7 |
| latency ms/icon, batch 1, p95 | 4064.1 |
| throughput ms/icon, batch 64 | 88.2 |
| model calls per icon (decoding) | 826.0 |
| torch / CUDA | 2.14.1+cu130 / 13.0 |
Latency covers encoding and decoding to a validated token program; it excludes rasterising the SVG. Samples: samples.png, rows are reference, model, nearest training icon.
State transitions
- running2026-10-13T18:02:53Z
- completed2026-10-13T19:04:19Z
Run record
Verbatim from runs/r2s-v10-img192-5247da7-bcf071a7-360aa736/run.yaml, the record committed before launch.
schema_version
2
run_id
r2s-v10-img192-5247da7-bcf071a7-360aa736
state
completed
planned_at
2026-10-13T18:02:53Z
hypothesis
With input renders 144 -> 192 px (image grid 18 x 18 -> 24 x 24), and everything else as in v9, the render-to-program model transcribes the 339 validation renders with lower mean 72 px pixel error than v9 trained on the same machine and seed. Pass requires the paired 95% bootstrap interval on the per-icon reduction against that v9 rerun to exclude zero. Its greedy latency on the RTX 4080 is recorded; a pass that is too slow for the browser demo is a teacher for distillation, not the demo model.
parent_run
r2s-full-v9-colour
git_commit
5247da7ce9cb85d7952d2cf5833ef2daa3a416a1
dirty_patch_sha256
79f30c7a2834c8d8…config
configs/render2svg/v10-img192.yaml
config_sha256
bcf071a717bd97ed…config_resolved
model
image_size
192
d_model
256
heads
8
encoder_layers
2
decoder_layers
6
feedforward
1024
dropout
0.1
metric
true
fourier
10
order
path
training
steps
60000
batch_size
32
learning_rate
0.0005
weight_decay
0.01
warmup_steps
1000
seed
7001
eval_every
2000
eval_icons
64
train_icons
None
evaluate_on_train
false
augment_variants
0
augment_seed
9001
augment_mirror
0.5
augment_max_shift
48
augment_colour
0.1
augment_online
true
compose_probability
0.5
compose_parts
4
augment_original
0.1
loader_workers
16
extra_training
bf16
true
final_eval_icons
339
clip_icons
32
dataset
pilot_config
configs/learning/openmoji-g1-geometric-gate-v16.yaml
cache_sha256
360aa73657ed7ac2…train_icons
2681
augmented_variants
0
extra_training
None
augmented_sha256
None
evaluated_on
primary/validation
seed
7001
determinism
seeded; cuDNN and SDPA kernels not forced deterministic
model_parameters
9090338
command
.venv/bin/python -m mojidiff.learning.render2svg --config configs/render2svg/v10-img192.yaml
outputs
run_dir
runs/r2s-v10-img192-5247da7-bcf071a7-360aa736
checkpoints
/data/fs201459/cp98556/predictable/cache/runs/r2s-v10-img192-5247da7-bcf071a7-360aa736
notes
completed_at
2026-10-13T19:04:19Z
resource
device
NVIDIA H100
torch_version
2.14.1+cu130
cuda_version
13.0
python_version
3.12.14
peak_vram_gib
4.671260833740234
train_seconds
3430.5903772809543
inference_ms_per_icon
3652.6720445253886
inference_p95_ms_per_icon
4064.1099580097944
icons_per_second
0.2737721831607621
batched_ms_per_icon
88.23204529835493
batched_icons_per_call
64
excludes
rendering the output SVG to pixels
baselines
blank_canvas_pixel_error
0.1720480552069557
nearest_training_icon_pixel_error
0.08973124639576185
gate_n_omnisvg_zero_shot_top1
0.609375
nearest_training_icon_clip_top1
0.34375
result
icons
339
model_pixel_error
0.07779111817594411, 0.07229561135918146, 0.08325007181065651
model_pixel_error_median
0.07389352470636368
exact_program_rate
0.02064896755162242
rendered_rate
1.0
blank_pixel_error
0.1720480552069557
selected_step
50000
nearest_training_icon_pixel_error
0.08973124639576185, 0.0848205749792233, 0.09466393787711656
model_minus_baseline_error_reduction
0.011940128219817742, 0.007749849801869065, 0.016091551545820967
icons_model_beats_baseline
217
clip_retrieval
model_greedy
top1_rate
0.65625
top5_rate
0.84375
mean_rank
3.34375
clip_to_reference_mean
0.9224974792450666
chance_top1
0.03125
icons
32
nearest_training_icon
top1_rate
0.34375
top5_rate
0.59375
mean_rank
7.4375
clip_to_reference_mean
0.8629764690995216
chance_top1
0.03125
icons
32
gate_n_omnisvg_zero_shot_top1
0.609375
gate_n_omnisvg_best_of_12_top1
0.828125
model_calls_per_icon
826
predeclared_outcome
criterion
paired 95% bootstrap interval on the per-icon reduction against the v9 rerun excludes zero in favour
against
r2s-full-v9-colour-989aea8-b26ad95e-47646604
paired
icons
339
candidate_pixel_error
0.07779111817594411, 0.07239674141480335, 0.0834737572942891
parent_pixel_error
0.07241242935923602, 0.06769235660214266, 0.07717277473913198
parent_minus_candidate
-0.005378688816708102, -0.009184582231185857, -0.0016813316117843685
candidate_better_icons
142
parent_better_icons
184
interval_excludes_zero_in_favour
false
passed
false
rtx4080_latency
not measured yet