The captcha__x family of the ElliotVL supervised-fine-tuning pool, after VLM cleaning.
images
87,346
QA turns
283,846
answers rewritten by the cleaning pass
28,810
QA created by the cleaning pass (new_qa)
196,495 (69.2%)
shards
2
How this was cleaned
A vision-language model read each image together with its QA and judged the item. The pass is
not a filter that only removes rows — it rewrites answers it finds wrong but salvageable,
drops what it cannot salvage, and adds QA where the image supports more than the source provided.
Each row carries the judge's own record in clean_meta, including the cleaner identity, the
policy it applied, and its per-item scores for legibility, richness and coverage.
A large share of the QA here was written by the cleaning pass, not by the original dataset.
Across the pool that share runs from roughly half to over 80% of a family's turns, and it is
reported in the table above. Those rows have an empty org_answer because no pre-clean original
exists. Treat them as model-generated supervision: they were judged against the image, but they
are not human annotation, and model-written QA is where formatting defects are most likely.
The effect on the answers that were carried over is substantive rather than cosmetic. In this pool the pass corrects values, not just
wording — a curve's critical point restated from 4.00 to 2.00, a computed ratio from 1 to
Approximately 1.33 — and for some families it removed the content entirely, which is why those
families are absent here rather than published empty.
answer vs org_answer
answer — the cleaned answer. Train on this.
org_answer — the pre-cleaning answer from the same (image, question) in the uncleaned pool.
It is empty for QA the cleaning pass added, which have no original.
Where the two differ, the difference is the correction. Keeping both makes every edit auditable
instead of asking you to take the cleaning on trust.
Schema
image — HF Image(), renders directly in the dataset viewer
image_sha256 — content hash of the image
n_qa — number of turns attached to this image
qa — list of {question, answer, org_answer}, in source order
family / clean_meta — provenance and the judge's per-row record
A row with more than one entry in qa is a conversation over a single image, not a set of
independent questions — the turns depend on each other and the image belongs to all of them.
Keep them together and in order when training.
Parquet row groups are written at ~32 MB so the dataset viewer pages without stalling.
What is NOT claimed
No rejection sampling was run against this family: there is no accept/reject verdict per row, and
answer is the cleaned reference rather than a model generation that a string verifier confirmed.
For that, use the *_RS_think / *_rejected datasets in elliot-mllm.
Examples
10 rows taken straight from this dataset. Images are the original files as stored here - no downscaling, no recompression - and every turn is shown in full.
Example 1 — 3 turns on one image
#
question
answer (cleaned)
org_answer (pre-judge)
1
What is the text shown in this captcha image?
ZWXGC
Zwxgc
2
What graphical elements are overlaid across the text?
Two thin, wavy black lines.
(unchanged)
3
What is the background color of the image?
Light pink.
(unchanged)
Example 2 — 3 turns on one image
#
question
answer (cleaned)
org_answer (pre-judge)
1
Can you tell me what letters and numbers are visible in this captcha?
gBGI1
gDGTi
2
What visual interference is present over the text?
Wavy lines, one green and one black, cross through the characters.
(unchanged)
3
What is the background color of the image?
Grey
(unchanged)
Example 3 — 4 turns on one image
#
question
answer
1
Can you tell me what letters and numbers are visible in this captcha?
hYI61
2
Describe the lines crossing the text.
A green diagonal line and white jagged lines cross over the text.
3
What is the background color?
Dark purple.
4
What is the color of the text characters?
Light pink.
Example 4 — 3 turns on one image
#
question
answer (cleaned)
org_answer (pre-judge)
1
Please transcribe the text content from this captcha verification image.
4RUNZ
4Ruvz
2
What is the background color of the image?
Black
(unchanged)
3
What color are the text characters?
Light purple
(unchanged)
Example 5 — 3 turns on one image
#
question
answer
1
What is the text shown in this captcha image?
QdLhX
2
What is the background color of the image?
Dark green
3
What colors are the lines crossing through the text?
Reddish-brown and green
Example 6 — 3 turns on one image
#
question
answer
1
What is the alphanumeric code shown in this image?
JCdsC
2
What color are the letters in the alphanumeric code?
Green
3
What graphical elements are superimposed over the text?
Wavy blue and green lines
Example 7 — 3 turns on one image
#
question
answer
1
Please transcribe the text content from this captcha verification image.
X3FCF
2
What color are the characters in the captcha?
Green
3
Describe the visual noise overlaid on the text.
Reddish-brown strikethrough lines crossing horizontally through the characters.
Example 8 — 3 turns on one image
#
question
answer (cleaned)
org_answer (pre-judge)
1
Please read and identify the characters displayed in this verification code image.
YCSTV
ycsTV
2
What are the colors of the text and the background?
The text is light purple and the background is dark green.
(unchanged)
3
Describe the interference lines visible in the image.
There is a thin red line and a thicker green line crossing through the characters.
(unchanged)
Example 9 — 4 turns on one image
#
question
answer
1
What is the text shown in this captcha image?
g2j1k
2
What color is the text in the image?
Blue
3
What graphical element crosses horizontally through the text?
A wavy line
4
What is the background color of the image?
Black
Example 10 — 4 turns on one image
#
question
answer (cleaned)
org_answer (pre-judge)
1
What is the text shown in this captcha image?
Kh1AD
kh1AD
2
What is the color of the text in the image?
Dark teal or blue-green
(unchanged)
3
What is the background color of the image?
Pale yellow or beige
(unchanged)
4
Describe the visual noise or distortion present on the text.
A thin, slightly lighter blue line crosses horizontally through the middle of the characters.