wordart_cleaned

The wordart family of the ElliotVL supervised-fine-tuning pool, after VLM cleaning.

images	4,765
QA turns	12,859
answers rewritten by the cleaning pass	41
QA created by the cleaning pass (new_qa)	12,738 (99.1%)
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
1	What is the shape of the blue emblem?	Oval
2	What is visible directly below the emblem?	A black car grille
3	What is the background color of the emblem?	Dark blue

Example 2 — 4 turns on one image

#	question	answer
1	What words can you see in these images in sequence, separated by a semicolon?	RACCOON; e
2	What animal graphic is partially visible above the word 'RACCOON'?	A raccoon's face, specifically showing the left eye and forehead area.
3	What visual effect is present on the white print?	The print has a holographic or iridescent sheen.
4	Describe the font style of the word 'RACCOON'.	It is a white, arched, serif font with a drop shadow or outline effect.

Example 3 — 3 turns on one image

#	question	answer
1	What is the color of the text?	Dark red
2	What is the background color?	Beige
3	What style of font is used for the word?	Cursive script

Example 4 — 3 turns on one image

#	question	answer
1	What color are the letters on the sign?	Pink with a gold outline
2	Describe the background surface behind the sign.	Reddish-brown horizontal siding
3	What is the only complete word visible in the image?	Red

Example 5 — 2 turns on one image

#	question	answer
1	What font style is used for the text in the image?	Serif
2	What is the color of the text?	Dark brown

Example 6 — 3 turns on one image

#	question	answer
1	What is the main text written in the large, silver font?	Day!
2	What is the texture and color of the lettering?	Silver and glittery
3	What animal feature is partially visible at the bottom center of the image?	Reindeer antlers

Example 7 — 4 turns on one image

#	question	answer
1	What color are the letters in the image?	Maroon
2	What style of font is used for the word 'you'?	Cursive script
3	What is the background color and texture?	Brown textured surface
4	What visual effect gives the text a 3D appearance?	Drop shadow

Example 8 — 3 turns on one image

#	question	answer
1	What is the single word written in the image?	Boldly
2	What is the background color of the image?	Light blue
3	Describe the font style used for the text.	Black brush script

Example 9 — 2 turns on one image

#	question	answer
1	What is the background color of the image?	Salmon pink
2	What style of font is used for the text?	Handwritten cursive script

Example 10 — 3 turns on one image

#	question	answer
1	What sequence of uppercase letters is visible in the image?	ABCDEFGHIJKLMN
2	What are the colors of the text and the background?	The text is red or orange-red, and the background is a light beige or cream color.
3	Describe the style of the font used for the letters.	The letters are tall, condensed, bold, and sans-serif.