r2s-cur-rl-smoke-532c2d4-6eb59a8f-47646604
completed —
Hypothesis
Smoke test: render-reward fine-tuning from the final model runs end to end (6 steps, 4 icons x 4 samples on the 16 GiB card).
Visual output
Written result
Hypothesis. Smoke test: render-reward fine-tuning from the final model runs end to end (6 steps, 4 icons x 4 samples on the 16 GiB card).
Result
| measure | value |
|---|---|
| icons evaluated | 16 (primary/validation) |
| model pixel error, mean [95% CI] | 0.0642 [0.0456, 0.0830] |
| model pixel error, median | 0.0606 |
| exact program rate | 0.000 |
| rendered rate | 1.000 |
| blank canvas pixel error | 0.1829 |
| nearest training icon pixel error | 0.0775 [0.0569, 0.1018] |
| error reduction vs nearest icon | 0.0133 [-0.0093, 0.0349] |
| icons where model beats nearest icon | 11 |
| CLIP top-1, model greedy (4 Gate N icons) | 1.000 |
| CLIP top-1, nearest training icon | 0.250 |
| CLIP top-1, OmniSVG 4B zero-shot (Gate N) | 0.609 |
Resources
| measure | value |
|---|---|
| device | NVIDIA GeForce RTX 4080 |
| parameters | 9,025,826 |
| train seconds | 21 |
| peak VRAM GiB | 2.6904897689819336 |
| latency ms/icon, batch 1, median | 1266.5 |
| latency ms/icon, batch 1, p95 | 1268.6 |
| throughput ms/icon, batch 16 | 122.4 |
| model calls per icon (decoding) | 621.0 |
| torch / CUDA | 2.14.0a0+4fdf77b940.nv26.08 / 13.4 |
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
- running2026-10-31T17:39:31Z
- completed2026-10-31T17:40:25Z
Run record
Verbatim from runs/r2s-cur-rl-smoke-532c2d4-6eb59a8f-47646604/run.yaml, the record committed before launch.
schema_version
2
run_id
r2s-cur-rl-smoke-532c2d4-6eb59a8f-47646604
state
completed
planned_at
2026-10-31T17:39:31Z
hypothesis
Smoke test: render-reward fine-tuning from the final model runs end to end (6 steps, 4 icons x 4 samples on the 16 GiB card).
parent_run
r2s-cur-d02-s120k-seed2-53b3108-46294452-47646604
git_commit
532c2d4782562261abb60738a7997ab3d4bbcc86
dirty_patch_sha256
a951c3b97bef875e…config
configs/render2svg/cur-rl-smoke.yaml
config_sha256
6eb59a8f38d83b7d…config_resolved
model
image_size
144
d_model
256
heads
8
encoder_layers
2
decoder_layers
6
feedforward
1024
dropout
0.2
metric
true
fourier
10
order
path
sources
0
training
steps
6
batch_size
4
learning_rate
2e-05
weight_decay
0.01
warmup_steps
2
seed
7001
eval_every
3
eval_icons
16
train_icons
None
evaluate_on_train
false
augment_variants
0
augment_seed
9001
augment_mirror
0.5
augment_max_shift
48
augment_colour
0.5
augment_online
false
compose_probability
0.0
compose_parts
4
augment_original
0.1
loader_workers
10
extra_training
degrade_probability
0.0
degrade_kinds
lineart, sloppy, flat
degrade_eval_seed
0
init_run
r2s-cur-d02-s120k-seed2-53b3108-46294452-47646604
init_sha256
38daeca4dd27fb4c…rl_samples
4
rl_temperature
1.0
rl_anchor
0.5
max_seconds
None
bf16
true
final_eval_icons
16
clip_icons
4
dataset
pilot_config
configs/learning/openmoji-g1-geometric-gate-v16.yaml
cache_sha256
476466042da98d72…train_icons
2681
augmented_variants
0
extra_training
None
augmented_sha256
None
evaluated_on
primary/validation
evaluation_inputs
clean renders
init
run
r2s-cur-d02-s120k-seed2-53b3108-46294452-47646604
checkpoint
/home/dev/.cache/mojidiff/runs/r2s-cur-d02-s120k-seed2-53b3108-46294452-47646604/best.pt
sha256
38daeca4dd27fb4c…step
64000
seed
7001
determinism
seeded; cuDNN and SDPA kernels not forced deterministic
model_parameters
9025826
command
.venv/bin/python -m mojidiff.learning.render2svg --config configs/render2svg/cur-rl-smoke.yaml
outputs
run_dir
runs/r2s-cur-rl-smoke-532c2d4-6eb59a8f-47646604
checkpoints
/home/dev/.cache/mojidiff/runs/r2s-cur-rl-smoke-532c2d4-6eb59a8f-47646604
notes
stopped_at_step
6
completed_at
2026-10-31T17:40:25Z
resource
device
NVIDIA GeForce RTX 4080
torch_version
2.14.0a0+4fdf77b940.nv26.08
cuda_version
13.4
python_version
3.12.3
peak_vram_gib
2.6904897689819336
train_seconds
20.500978632015176
inference_ms_per_icon
1266.5482920128852
inference_p95_ms_per_icon
1268.6366389971226
icons_per_second
0.7895474703224553
batched_ms_per_icon
122.37189293955453
batched_icons_per_call
16
excludes
rendering the output SVG to pixels
baselines
blank_canvas_pixel_error
0.18288448872044683
nearest_training_icon_pixel_error
0.07745827548205853
gate_n_omnisvg_zero_shot_top1
0.609375
nearest_training_icon_clip_top1
0.25
result
icons
16
model_pixel_error
0.06416897277813405, 0.04558524929743726, 0.0830126314162044
model_pixel_error_median
0.06061196327209473
exact_program_rate
0.0
rendered_rate
1.0
blank_pixel_error
0.18288448872044683
selected_step
6
nearest_training_icon_pixel_error
0.07745827548205853, 0.05690388170769438, 0.10184130002744496
model_minus_baseline_error_reduction
0.013289302703924477, -0.009265280261752195, 0.034862902690656485
icons_model_beats_baseline
11
clip_retrieval
model_greedy
top1_rate
1.0
top5_rate
1.0
mean_rank
1.0
clip_to_reference_mean
0.9236414879560471
chance_top1
0.25
icons
4
nearest_training_icon
top1_rate
0.25
top5_rate
1.0
mean_rank
2.0
clip_to_reference_mean
0.8300648927688599
chance_top1
0.25
icons
4
gate_n_omnisvg_zero_shot_top1
0.609375
gate_n_omnisvg_best_of_12_top1
0.828125
model_calls_per_icon
621