r2s-cur-d02-sh64-s1-e5d1296-78ba2ddc-47646604
completed —
Hypothesis
With exact translations of up to 16 view units instead of 12 and everything else as dropout 0.2 at 60k steps, the model transcribes the 339 validation renders better than the dropout-0.2 run of the same seed. Judged over seeds 7001-7008: the mean seed-matched difference with a bootstrap interval over seeds excluding zero.
Visual output
Written result
Hypothesis. With exact translations of up to 16 view units instead of 12 and everything else as dropout 0.2 at 60k steps, the model transcribes the 339 validation renders better than the dropout-0.2 run of the same seed. Judged over seeds 7001-7008: the mean seed-matched difference with a bootstrap interval over seeds excluding zero.
Result
| measure | value |
|---|---|
| icons evaluated | 339 (primary/validation) |
| model pixel error, mean [95% CI] | 0.0655 [0.0606, 0.0706] |
| model pixel error, median | 0.0603 |
| exact program rate | 0.032 |
| 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 | 0.0242 [0.0201, 0.0284] |
| icons where model beats nearest icon | 267 |
| 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 B300 SXM6 AC |
| parameters | 9,025,826 |
| train seconds | 2620 |
| peak VRAM GiB | 4.388660430908203 |
| latency ms/icon, batch 1, median | 3187.3 |
| latency ms/icon, batch 1, p95 | 3210.5 |
| throughput ms/icon, batch 64 | 113.8 |
| model calls per icon (decoding) | 647.0 |
| 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
- running2026-10-31T08:23:48Z
- completed2026-10-31T09:09:44Z
Run record
Verbatim from runs/r2s-cur-d02-sh64-s1-e5d1296-78ba2ddc-47646604/run.yaml, the record committed before launch.
schema_version
2
run_id
r2s-cur-d02-sh64-s1-e5d1296-78ba2ddc-47646604
state
completed
planned_at
2026-10-31T08:23:48Z
hypothesis
With exact translations of up to 16 view units instead of 12 and everything else as dropout 0.2 at 60k steps, the model transcribes the 339 validation renders better than the dropout-0.2 run of the same seed. Judged over seeds 7001-7008: the mean seed-matched difference with a bootstrap interval over seeds excluding zero.
parent_run
r2s-cur-dropout02
git_commit
e5d12964522dc69a79ff684b7af1b347864658ce
dirty_patch_sha256
config
configs/render2svg/cur-d02-sh64-s1.yaml
config_sha256
78ba2ddcf35fcdc6…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
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
64
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
476466042da98d72…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
9025826
command
.venv/bin/python -m mojidiff.learning.render2svg --config configs/render2svg/cur-d02-sh64-s1.yaml
outputs
run_dir
runs/r2s-cur-d02-sh64-s1-e5d1296-78ba2ddc-47646604
checkpoints
/home/eai-jdkxz/eai-team11/mojidiff/cache/runs/r2s-cur-d02-sh64-s1-e5d1296-78ba2ddc-47646604
notes
completed_at
2026-10-31T09:09:44Z
resource
device
NVIDIA B300 SXM6 AC
torch_version
2.14.1+cu130
cuda_version
13.0
python_version
3.12.3
peak_vram_gib
4.388660430908203
train_seconds
2620.0872386349947
inference_ms_per_icon
3187.2755289950874
inference_p95_ms_per_icon
3210.487043019384
icons_per_second
0.31374758501512073
batched_ms_per_icon
113.76695285980531
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
0.06553029087396055, 0.06058827232093383, 0.0705721005835652
model_pixel_error_median
0.06029361113905907
exact_program_rate
0.032448377581120944
rendered_rate
1.0
blank_pixel_error
0.1720480552069557
selected_step
52000
nearest_training_icon_pixel_error
0.08973124639576185, 0.0848205749792233, 0.09466393787711656
model_minus_baseline_error_reduction
0.024200955521801303, 0.0200664485543332, 0.028353994311612528
icons_model_beats_baseline
267
clip_retrieval
model_greedy
top1_rate
0.65625
top5_rate
0.96875
mean_rank
2.03125
clip_to_reference_mean
0.9366293549537659
chance_top1
0.03125
icons
32
nearest_training_icon
top1_rate
0.34375
top5_rate
0.59375
mean_rank
7.40625
clip_to_reference_mean
0.8630617223680019
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
647