Datasets:

Elliot-Data
/

crohme_train_cleaned

---
license: other
task_categories:
- visual-question-answering
language:
- en
---
# crohme_train_cleaned

The **crohme_train** family of the ElliotVL supervised-fine-tuning pool, **after VLM cleaning**.
| | |
|---|---|
| images | 8,708 |
| QA turns | 22,610 |
| answers rewritten by the cleaning pass | 790 |
| QA created by the cleaning pass (`new_qa`) | 15,140 (67.0%) |
| 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](https://huggingface.co/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

![case1](samples/case1.png)

| # | question | answer |
|---|---|---|
| 1 | Recognize the handwritten mathematical expression. Return only the expression transcription. | 1 |
| 2 | What is the background color of the image? | Black |
| 3 | What color is the digit? | White |


### Example 2 — 4 turns on one image

![case2](samples/case2.png)

| # | question | answer |
|---|---|---|
| 1 | Recognize the handwritten mathematical expression. Return only the expression transcription. | 1 |
| 2 | What is the background color of the image? | Black |
| 3 | What is the color of the character? | White |
| 4 | Does the character appear to be in a serif or sans-serif font? | Serif |


### Example 3 — 4 turns on one image

![case3](samples/case3.png)

| # | question | answer |
|---|---|---|
| 1 | Recognize the handwritten mathematical expression. Return only the expression transcription. | 1 |
| 2 | What is the background color of the image? | Black |
| 3 | What is the color of the digit? | White |
| 4 | Where is the digit positioned in the frame? | Centered |


### Example 4 — 3 turns on one image

![case4](samples/case4.png)

| # | question | answer |
|---|---|---|
| 1 | Recognize the handwritten mathematical expression. Return only the expression transcription. | l |
| 2 | What are the colors of the character and the background? | The character is white and the background is black. |
| 3 | Where is the character positioned in the image? | The character is centered in the image. |


### Example 5

![case5](samples/case5.png)

| # | question | answer |
|---|---|---|
| 1 | What character is displayed in the image? | E |


### Example 6 — 3 turns on one image

![case6](samples/case6.png)

| # | question | answer (cleaned) | org_answer (pre-judge) |
|---|---|---|---|
| 1 | Recognize the handwritten mathematical expression. Return only the expression transcription. | \sum x - \frac { \beta } { k } \neq \frac { T - 0 } { y ^ { e } ( \phi ) } | \sum x - \frac { \beta } { k } \neq \frac { T - o } { y ^ { e } ( \phi ) } |
| 2 | What is the visual style of the text in the image? | Handwritten white text on a black background. | _(unchanged)_ |
| 3 | What type of mathematical relationship is depicted in the equation? | An inequality. | _(unchanged)_ |


### Example 7 — 2 turns on one image

![case7](samples/case7.png)

| # | question | answer |
|---|---|---|
| 1 | Recognize the handwritten mathematical expression. Return only the expression transcription. | 1 + 2 + \cdots + n = \frac { n ( n + 1 ) } { 2 } |
| 2 | What are the colors of the text and the background in the image? | White text on a black background. |


### Example 8 — 2 turns on one image

![case8](samples/case8.png)

| # | question | answer |
|---|---|---|
| 1 | Recognize the handwritten mathematical expression. Return only the expression transcription. | \frac { \sqrt { 1 + \cos ^ { 2 } ( \theta ) } } { 2 } = \cos ( 2 \theta ) |
| 2 | Describe the visual style of the text and background in the image. | The image shows white handwritten text on a black background. |


### Example 9 — 4 turns on one image

![case9](samples/case9.png)

| # | question | answer |
|---|---|---|
| 1 | Recognize the handwritten mathematical expression. Return only the expression transcription. | u _ { 1 } = e ^ { \frac { 2 i \pi } { 9 } } |
| 2 | What is the background color of the image? | Black |
| 3 | What is the color of the mathematical expression? | White |
| 4 | Is the text in the image printed or handwritten? | Handwritten |


### Example 10 — 3 turns on one image

![case10](samples/case10.png)

| # | question | answer |
|---|---|---|
| 1 | Recognize the handwritten mathematical expression. Return only the expression transcription. | \int x ^ { n } d x = \frac { x ^ { n + 1 } } { n + 1 } + C |
| 2 | What are the colors of the text and the background in the image? | The text is white and the background is black. |
| 3 | What style of writing is used for the mathematical expression? | The expression is written in a white, handwritten or digital marker style. |

	---
	license: other
	task_categories:
	- visual-question-answering
	language:
	- en
	---

	# crohme_train_cleaned

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

	\| \| \|
	\|---\|---\|
	\| images \| 8,708 \|
	\| QA turns \| 22,610 \|
	\| answers rewritten by the cleaning pass \| 790 \|
	\| QA created by the cleaning pass (`new_qa`) \| 15,140 (67.0%) \|
	\| 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](https://huggingface.co/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

	![case1](samples/case1.png)

	\| # \| question \| answer \|
	\|---\|---\|---\|
	\| 1 \| Recognize the handwritten mathematical expression. Return only the expression transcription. \| 1 \|
	\| 2 \| What is the background color of the image? \| Black \|
	\| 3 \| What color is the digit? \| White \|


	### Example 2 — 4 turns on one image

	![case2](samples/case2.png)

	\| # \| question \| answer \|
	\|---\|---\|---\|
	\| 1 \| Recognize the handwritten mathematical expression. Return only the expression transcription. \| 1 \|
	\| 2 \| What is the background color of the image? \| Black \|
	\| 3 \| What is the color of the character? \| White \|
	\| 4 \| Does the character appear to be in a serif or sans-serif font? \| Serif \|


	### Example 3 — 4 turns on one image

	![case3](samples/case3.png)

	\| # \| question \| answer \|
	\|---\|---\|---\|
	\| 1 \| Recognize the handwritten mathematical expression. Return only the expression transcription. \| 1 \|
	\| 2 \| What is the background color of the image? \| Black \|
	\| 3 \| What is the color of the digit? \| White \|
	\| 4 \| Where is the digit positioned in the frame? \| Centered \|


	### Example 4 — 3 turns on one image

	![case4](samples/case4.png)

	\| # \| question \| answer \|
	\|---\|---\|---\|
	\| 1 \| Recognize the handwritten mathematical expression. Return only the expression transcription. \| l \|
	\| 2 \| What are the colors of the character and the background? \| The character is white and the background is black. \|
	\| 3 \| Where is the character positioned in the image? \| The character is centered in the image. \|


	### Example 5

	![case5](samples/case5.png)

	\| # \| question \| answer \|
	\|---\|---\|---\|
	\| 1 \| What character is displayed in the image? \| E \|


	### Example 6 — 3 turns on one image

	![case6](samples/case6.png)

	\| # \| question \| answer (cleaned) \| org_answer (pre-judge) \|
	\|---\|---\|---\|---\|
	\| 1 \| Recognize the handwritten mathematical expression. Return only the expression transcription. \| \sum x - \frac { \beta } { k } \neq \frac { T - 0 } { y ^ { e } ( \phi ) } \| \sum x - \frac { \beta } { k } \neq \frac { T - o } { y ^ { e } ( \phi ) } \|
	\| 2 \| What is the visual style of the text in the image? \| Handwritten white text on a black background. \| _(unchanged)_ \|
	\| 3 \| What type of mathematical relationship is depicted in the equation? \| An inequality. \| _(unchanged)_ \|


	### Example 7 — 2 turns on one image

	![case7](samples/case7.png)

	\| # \| question \| answer \|
	\|---\|---\|---\|
	\| 1 \| Recognize the handwritten mathematical expression. Return only the expression transcription. \| 1 + 2 + \cdots + n = \frac { n ( n + 1 ) } { 2 } \|
	\| 2 \| What are the colors of the text and the background in the image? \| White text on a black background. \|


	### Example 8 — 2 turns on one image

	![case8](samples/case8.png)

	\| # \| question \| answer \|
	\|---\|---\|---\|
	\| 1 \| Recognize the handwritten mathematical expression. Return only the expression transcription. \| \frac { \sqrt { 1 + \cos ^ { 2 } ( \theta ) } } { 2 } = \cos ( 2 \theta ) \|
	\| 2 \| Describe the visual style of the text and background in the image. \| The image shows white handwritten text on a black background. \|


	### Example 9 — 4 turns on one image

	![case9](samples/case9.png)

	\| # \| question \| answer \|
	\|---\|---\|---\|
	\| 1 \| Recognize the handwritten mathematical expression. Return only the expression transcription. \| u _ { 1 } = e ^ { \frac { 2 i \pi } { 9 } } \|
	\| 2 \| What is the background color of the image? \| Black \|
	\| 3 \| What is the color of the mathematical expression? \| White \|
	\| 4 \| Is the text in the image printed or handwritten? \| Handwritten \|


	### Example 10 — 3 turns on one image

	![case10](samples/case10.png)

	\| # \| question \| answer \|
	\|---\|---\|---\|
	\| 1 \| Recognize the handwritten mathematical expression. Return only the expression transcription. \| \int x ^ { n } d x = \frac { x ^ { n + 1 } } { n + 1 } + C \|
	\| 2 \| What are the colors of the text and the background in the image? \| The text is white and the background is black. \|
	\| 3 \| What style of writing is used for the mathematical expression? \| The expression is written in a white, handwritten or digital marker style. \|