omnisvg-m1-finetune-greedy-19bbf5b-32icons-9b9b1699
completed —
Measured result
Verbatim from the run's summary.json.
checks
clip_gain_over_control
false
ended_rate
false
reference_top1_rate
false
clip
repo
openai/clip-vit-base-patch32
revision
3d74acf9a28c67741b2f4f2ea7635f0aaf6f0268
clip_to_caption
mean
0.22422297205775976
median
0.21936502307653427
n
32
clip_to_reference
mean
0.8268364705145359
median
0.8249839246273041
n
32
codec_valid_rate
0.53125
config_sha256
7ac1e8c59efa7cd8…control
clip_to_reference_mean
0.8268364705145359
control_clip_to_reference_mean
0.8112438190728426
control_codec_valid_rate
0.734375
control_reference_top1_rate
0.078125
control_rows_sha256
3f6131b329d040a5…gain_ci95
-0.0028321677586063744, 0.033771081687882544
icons_improved
21
mean_gain
0.015592651441693306
paired_icons
32
report_root
reports/learning/omnisvg-m1-control
criteria
clip_gain_over_control_ci_excludes_zero
true
min_ended_rate
0.9
min_reference_top1_rate
0.25
decoded_rate
1.0
drawings
32
ended_rate
0.5
failures
encode:program needs 1 contours and up to 511 segments
1
encode:program needs 115 contours and up to 9 segments
1
encode:program needs 24 contours and up to 172 segments
1
encode:program needs 50 contours and up to 12 segments
1
encode:program needs 52 contours and up to 11 segments
1
encode:program needs 58 contours and up to 10 segments
2
encode:program needs 60 contours and up to 15 segments
1
encode:program needs 7 contours and up to 458 segments
1
encode:program needs 71 contours and up to 11 segments
1
encode:program needs 8 contours and up to 411 segments
1
encode:program needs 91 contours and up to 11 segments
2
normalize:geometry_error: M has sets of 2 args, 1 invalid
1
pack:program needs 129 packed segments but capacity is 128
1
icons
32
max_new_tokens
2048
mean_snap_distance_rgb
5.037010170875593
median_paths
8.5
median_tokens
1261.5
model
adapter
data/processed/prior/omnisvg-m1-finetune/adapter.zip
adapter_sha256
23bf01ba394a1e04…checkpoint_sha256
39f36759fc677b32…fine_tuned
true
parameters
3846914048
repo
OmniSVG/OmniSVG1.1_4B
revision
117d4c4541839e02943554fe71c0a0d54fcef819
predeclared_outcome
falsified
prompt_style
training
reference_rank
chance_top1
0.03125
mean
11.4375
scored
32
top1_rate
0.0625
top5_rate
0.28125
rows_sha256
e3d92d4a0230ddd7…samples_per_icon
1
sampling
greedy
true
repetition_penalty
1.0
schema_version
1
sheet_sha256
e5599a6163b4d261…study_version
omnisvg-m1-finetune-greedy
timing
generate_seconds
688.0395654331078
load_seconds
37.51105879098759
peak_vram_gib
8.074963569641113
seconds_per_drawing
21.501236419784618
Visual output
State transitions
- planned2026-09-21T14:46:57Z
- completed2026-09-21T15:21:20Zgreedy decoding: 16 of 32 drawings loop to the cap, 8 are the blue square, the rest a bust, a yellow face, a triangle, an octagon; CLIP-to-reference 0.827, paired gain +0.016 with an interval of -0.003 to +0.034, 21 of 32 icons improved, the right icon first for 2 of 32, top five for 9
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-m1-finetune-greedy-19bbf5b-32icons-9b9b1699
config
configs/learning/omnisvg-m1-finetune-greedy.yaml
outputs
report_root
reports/learning/omnisvg-m1-finetune-greedy