MojiDiff

← experiments

omnisvg-m1-zeroshot-bcce3f0-32icons-9b9b1699

completed —

Measured result

Verbatim from the run's summary.json.

clip
repo
openai/clip-vit-base-patch32
revision
3d74acf9a28c67741b2f4f2ea7635f0aaf6f0268
clip_to_caption
mean
0.2187486533075571
median
0.2181549295783043
n
64
clip_to_reference
mean
0.8113202741369605
median
0.8110454678535461
n
64
codec_valid_rate
0.734375
config_sha256
8b9dcbc83439d223…
decoded_rate
1.0
drawings
64
ended_rate
0.765625
failures
encode:program needs 2 contours and up to 1003 segments
1
encode:program needs 2 contours and up to 504 segments
1
encode:program needs 3 contours and up to 493 segments
1
encode:program needs 3 contours and up to 494 segments
4
encode:program needs 3 contours and up to 496 segments
1
encode:program needs 4 contours and up to 380 segments
1
encode:program needs 5 contours and up to 442 segments
1
encode:program needs 6 contours and up to 475 segments
1
encode:program needs 6 contours and up to 894 segments
1
encode:program needs 99 contours and up to 11 segments
1
normalize:unsupported_color: fill color 'currentColor' is unsupported
1
pack:program needs 183 packed segments but capacity is 128
1
pack:program needs 222 packed segments but capacity is 128
1
pack:program needs 343 packed segments but capacity is 128
1
icons
32
max_new_tokens
2048
mean_snap_distance_rgb
24.37359129064963
median_paths
4.0
median_tokens
263.0
model
checkpoint_sha256
39f36759fc677b32…
fine_tuned
false
parameters
3846914048
repo
OmniSVG/OmniSVG1.1_4B
revision
117d4c4541839e02943554fe71c0a0d54fcef819
rows_sha256
f666bbac85fee3db…
samples_per_icon
2
schema_version
1
sheet_sha256
d7be9c5e326f1850…
study_version
omnisvg-m1-zeroshot
timing
generate_seconds
415.70064489819924
load_seconds
36.76467291099834
peak_vram_gib
8.023042678833008
seconds_per_drawing
6.495322576534363

Visual output

samples
samples.png

State transitions

  1. planned2026-09-21T13:03:36Z
  2. completed2026-09-21T13:25:35Z64 drawings from 32 held-out annotations, all decodable, 77% ended within 2,048 tokens, 73% taken by the codec (the rest need more than 128 segments or more than the bucket's contours); CLIP-to-reference 0.811, CLIP-to-caption 0.219

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-zeroshot-bcce3f0-32icons-9b9b1699
config
configs/learning/omnisvg-m1-zeroshot.yaml
outputs
report_root
reports/learning/omnisvg-m1-zeroshot