omnisvg-m1-control-trainprompt-52b1950-32icons-9b9b1699
completed —
Measured result
Verbatim from the run's summary.json.
clip
repo
openai/clip-vit-base-patch32
revision
3d74acf9a28c67741b2f4f2ea7635f0aaf6f0268
clip_to_caption
mean
0.2165887770242989
median
0.2167740911245346
n
64
clip_to_reference
mean
0.786869028583169
median
0.7867538928985596
n
64
codec_valid_rate
0.6875
config_sha256
7d9a7c2651aeec91…decoded_rate
1.0
drawings
64
ended_rate
0.8125
failures
encode:endpoint coordinate is outside 0..72 and clamping is disabled
3
encode:program needs 1 contours and up to 510 segments
1
encode:program needs 13 contours and up to 361 segments
1
encode:program needs 2 contours and up to 236 segments
1
encode:program needs 2 contours and up to 503 segments
3
encode:program needs 2 contours and up to 504 segments
1
encode:program needs 21 contours and up to 411 segments
1
encode:program needs 5 contours and up to 908 segments
1
encode:program needs 7 contours and up to 479 segments
1
encode:program needs 84 contours and up to 21 segments
1
normalize:geometry_error: M has sets of 2 args, 1 invalid
2
normalize:unsupported_color: fill color 'currentColor' is unsupported
3
pack:program needs 133 packed segments but capacity is 128
1
icons
32
max_new_tokens
2048
mean_snap_distance_rgb
12.827936416032223
median_paths
1.0
median_tokens
154.0
model
checkpoint_sha256
39f36759fc677b32…fine_tuned
false
parameters
3846914048
repo
OmniSVG/OmniSVG1.1_4B
revision
117d4c4541839e02943554fe71c0a0d54fcef819
prompt_style
training
reference_rank
chance_top1
0.03125
mean
15.859375
scored
64
top1_rate
0.046875
top5_rate
0.1875
rows_sha256
17d20f0342911475…samples_per_icon
2
schema_version
1
sheet_sha256
e7f0e4688042f8ca…study_version
omnisvg-m1-control-trainprompt
timing
generate_seconds
384.5285660540394
load_seconds
36.56999058101792
peak_vram_gib
8.019460678100586
seconds_per_drawing
6.008258844594366
Visual output
State transitions
- planned2026-09-21T13:33:27Z
- completed2026-09-21T14:15:48Zunder the trainer's prompt the released model draws shorter (median 154 tokens against 263), ends more often (81%), scores lower (CLIP-to-reference 0.787, top-1 0.047, mean rank 15.9)
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-control-trainprompt-52b1950-32icons-9b9b1699
config
configs/learning/omnisvg-m1-control-trainprompt.yaml
outputs
report_root
reports/learning/omnisvg-m1-control-trainprompt