omnisvg-n1-image-finetune-8664e9d-32icons-9b9b1699
completed —
Measured behaviour
Table view
| optimizer step | model |
|---|---|
| 0 | 2.5830 |
| 25 | 1.9420 |
| 50 | 1.3936 |
| 75 | 1.2960 |
| 100 | 1.2537 |
| 125 | 1.2270 |
| 150 | 1.2063 |
| 175 | 1.1894 |
| 200 | 1.1729 |
| 225 | 1.1640 |
| 250 | 1.1489 |
| 275 | 1.1410 |
| 300 | 1.1362 |
| 325 | 1.1347 |
| 350 | 1.1325 |
| 375 | 1.1416 |
| 400 | 1.1255 |
| 425 | 1.1283 |
| 450 | 1.1297 |
| 475 | 1.1195 |
| 500 | 1.1242 |
Measured result
Verbatim from the run's summary.json.
checks
clip_gain_over_control
false
ended_rate
false
memorisation
true
reference_top1_rate
false
clip
repo
openai/clip-vit-base-patch32
revision
3d74acf9a28c67741b2f4f2ea7635f0aaf6f0268
clip_to_caption
mean
0.23430775268934667
median
0.22995629161596298
n
64
clip_to_reference
mean
0.8672871282324195
median
0.8572697341442108
n
64
codec_valid_rate
0.125
config_sha256
e0928ffd029c66a7…control
clip_to_reference_mean
0.8672871282324195
control_clip_to_reference_mean
0.9046997493132949
control_codec_valid_rate
0.140625
control_reference_top1_rate
0.609375
control_rows_sha256
5cabd027e8c4e617…gain_ci95
-0.05842225186061114, -0.01654697686899454
icons_improved
8
mean_gain
-0.0374126210808754
paired_icons
32
report_root
reports/learning/omnisvg-n1-image-control
criteria
clip_gain_over_control_ci_excludes_zero
true
max_memorised_exactly
0
min_ended_rate
0.9
min_reference_top1_rate
0.75
decoded_rate
1.0
drawings
64
ended_rate
0.28125
failures
encode:program needs 1 contours and up to 506 segments
1
encode:program needs 1 contours and up to 512 segments
1
encode:program needs 1 contours and up to 514 segments
2
encode:program needs 1 contours and up to 515 segments
1
encode:program needs 1 contours and up to 516 segments
1
encode:program needs 1 contours and up to 561 segments
1
encode:program needs 1 contours and up to 782 segments
1
encode:program needs 1 contours and up to 790 segments
1
encode:program needs 1 contours and up to 930 segments
1
encode:program needs 122 contours and up to 11 segments
1
encode:program needs 14 contours and up to 389 segments
1
encode:program needs 16 contours and up to 346 segments
1
encode:program needs 18 contours and up to 353 segments
1
encode:program needs 2 contours and up to 1008 segments
1
encode:program needs 2 contours and up to 300 segments
1
encode:program needs 2 contours and up to 303 segments
1
encode:program needs 2 contours and up to 463 segments
1
encode:program needs 2 contours and up to 501 segments
2
encode:program needs 2 contours and up to 504 segments
1
encode:program needs 2 contours and up to 505 segments
1
encode:program needs 3 contours and up to 162 segments
1
encode:program needs 3 contours and up to 461 segments
1
encode:program needs 3 contours and up to 477 segments
1
encode:program needs 3 contours and up to 493 segments
1
encode:program needs 3 contours and up to 547 segments
1
encode:program needs 34 contours and up to 20 segments
1
encode:program needs 35 contours and up to 19 segments
1
encode:program needs 37 contours and up to 15 segments
1
encode:program needs 4 contours and up to 421 segments
1
encode:program needs 42 contours and up to 11 segments
1
encode:program needs 5 contours and up to 445 segments
1
encode:program needs 5 contours and up to 473 segments
1
encode:program needs 5 contours and up to 476 segments
1
encode:program needs 5 contours and up to 483 segments
1
encode:program needs 51 contours and up to 10 segments
1
encode:program needs 52 contours and up to 11 segments
2
encode:program needs 53 contours and up to 11 segments
1
encode:program needs 56 contours and up to 12 segments
1
encode:program needs 6 contours and up to 430 segments
1
encode:program needs 6 contours and up to 481 segments
1
encode:program needs 60 contours and up to 9 segments
1
encode:program needs 66 contours and up to 18 segments
1
encode:program needs 7 contours and up to 444 segments
1
encode:program needs 7 contours and up to 480 segments
1
encode:program needs 8 contours and up to 224 segments
1
encode:program needs 8 contours and up to 451 segments
1
normalize:geometry_error: M has sets of 2 args, 1 invalid
1
pack:program needs 152 packed segments but capacity is 128
1
pack:program needs 175 packed segments but capacity is 128
1
pack:program needs 194 packed segments but capacity is 128
1
pack:program needs 224 packed segments but capacity is 128
1
pack:program needs 246 packed segments but capacity is 128
1
pack:program needs 576 packed segments but capacity is 128
1
icons
32
max_new_tokens
2048
mean_snap_distance_rgb
4.194017768290177
median_paths
5.0
median_tokens
2048.0
memorised_exactly
0
model
checkpoint_sha256
39f36759fc677b32…fine_tuned
true
lora
alpha
32
dropout
0.05
rank
16
parameters
3876847616
repo
OmniSVG/OmniSVG1.1_4B
revision
117d4c4541839e02943554fe71c0a0d54fcef819
trainable_parameters
29933568
predeclared_outcome
falsified
prompt_style
image
reference_rank
chance_top1
0.03125
mean
7.59375
scored
64
top1_rate
0.390625
top5_rate
0.5625
rows_sha256
9f8b60589d7bcd05…samples_per_icon
2
sampling
repetition_penalty
1.05
temperature
0.3
top_k
50
top_p
0.9
schema_version
1
sheet_sha256
d846d82e7aff51a0…study_version
omnisvg-n1-image-finetune
timing
generate_seconds
2128.9485412650392
peak_vram_gib
9.308963775634766
seconds_per_drawing
33.26482095726624
training
adapter_bytes
119816085
adapter_sha256
15b8be9922c3a678…condition
image
excluded_selection_over_max_tokens
0
excluded_train_over_max_tokens
0
held_out_nll
1.1195139529553957
initial_held_out_nll
2.5830178730228406
load_seconds
37.49828114002594
max_tokens
2400
peak_vram_gib
9.308963775634766
selected_step
475
selection_encode_failures
0
selection_icons
32
sequences_per_step
8
steps_run
500
train_encode_failures
0
train_icons
2681
train_seconds
2369.083323122002
Visual output
State transitions
- planned2026-09-22T14:44:21Z
- completed2026-09-22T16:19:56Zheld-out likelihood 2.58 to 1.12 nats a token over 500 steps (40 min, 9.3 GiB), and the drawings are worse than zero-shot: 28% end (control 50%), median 2,048 tokens, the right icon first for 25 of 64 (0.39; control 0.61), similarity 0.867 (control 0.905), paired gain -0.037 with an interval of -0.058 to -0.017; colours snap closer (4.2 against 8.5) and no training icon is reproduced
Run record
This run has no run.yaml. What follows is the identity and configuration carried by its rows in state/runs.jsonl, the append-only registry.
run_id
omnisvg-n1-image-finetune-8664e9d-32icons-9b9b1699
config
configs/learning/omnisvg-n1-image-finetune.yaml
outputs
adapter
data/processed/prior/omnisvg-n1-image-finetune/adapter.zip
adapter_sha256
15b8be9922c3a678…report_root
reports/learning/omnisvg-n1-image-finetune