README.md

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

# svrd_cleaned

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

| | |
|---|---|
| images | 1,827 |
| QA turns | 11,163 |
| answers rewritten by the cleaning pass | 2,766 |
| QA created by the cleaning pass (`new_qa`) | 7,858 (70.4%) |
| shards | 4 |

## 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 — 5 turns on one image

![case1](samples/case1.png)

| # | question | answer (cleaned) | org_answer (pre-judge) |
|---|---|---|---|
| 1 | \nExtract all key-value pairs and important information from this document. | I can identify the following key information from this document:<br><br>**Key-Value Relationships:**<br>• 报验号: NO.142002015001209<br>• 日期: 2015-12-30<br>• Tel: (0635)8506063<br>• Fax: (0635)8506069<br>• 页数: 第1页 共1页<br>• 地址: 山东省高唐县姜店工业开发区105国道<br>• 委托方: 高唐华特威科技有限公司<br>• 样品数量: 1个(3袋)<br>• 样品名称: 玉米芯颗粒粉<br>• 样品特征与状态: 符合要求 | I can identify the following key information from this document:\n\n**Key-Value Relationships:**\n• 报验号:: N0.142002015001209\n• 日期: 2015-12-30\n• Tel:: (0635)8506063\n• Fax:: (0635)8506069\n• 页数: 第1页 共1页\n• 地: 山东省高唐县姜店工业开发区105国\n• 委托方:: 高唐华特威科技有限公司\n• 样品数量:: 1个(3袋)\n• 样品名称:: 玉米芯颗粒粉\n• 样品特征与状态:: 符合要求 |
| 2 | What is the full name of the issuing organization and the document title? | Issuing Organization: 聊城检验检疫局技术中心 (LIAOCHENG ENTRY-EXIT INSPECTION AND QUARANTINE BUREAU TECHNOLOGY CENTER). Document Title: 检测报告 (TEST REPORT). | _(unchanged)_ |
| 3 | What are the sample number, sample receipt date, and the test result for 'Lead' (铅)? | Sample Number: 142015007939. Sample Receipt Date: 2015-12-24. Lead (铅) Result: 0.17 mg/kg. | _(unchanged)_ |
| 4 | What are the test methods used for 'Total Arsenic' (总砷) and 'Salmonella' (沙门氏菌)? | Total Arsenic: GB/T 5009.11-2003(第一法). Salmonella: GB/T 13091-2002. | _(unchanged)_ |
| 5 | What is the text in the 'Statement' (声明) section at the bottom of the document? | 1. 未经本实验室书面同意，不得部分地复制本报告。 2. 检验结果仅对受检样品有效。 | _(unchanged)_ |


### Example 2 — 6 turns on one image

![case2](samples/case2.png)

| # | question | answer |
|---|---|---|
| 1 | \nWhat are the key details and their values in this document? | I can identify the following key information from this document:\n\n**Key-Value Relationships:**\n• 资质有效期至:: 2021.12.13\n• 报告编号:: BMILP-NVTC-2019-033-2\n• 资质认定证书编号: 150121340108\n• 产品名称:: XYG-001阻性管道式消声器\n• 委托方:: 靖江市万春空调设备制造有限公司\n• 委托方地址:: 靖江市斜桥镇江平路夏仕港桥南首50米\n• 检测类别:: 委托\n• 签发日期:: 2019年6月17日\n• 有效日期:: 2022年6月16日\n• 联系电话:: 010-63524194 |
| 2 | What is the code located in the top right corner of the document? | BMILP-4-303 |
| 3 | What is the full title of the report displayed in the center? | 阻性管式消声器 检测报告 |
| 4 | Which organization issued this report, as indicated by the text and stamp at the bottom? | 北京市劳动保护科学研究所 |
| 5 | What is the page number information at the bottom of the document? | 第 1 页 共 5 页 |
| 6 | What are the address, zip code, and email listed in the footer? | Address: 北京市陶然亭路55号, Zip: 100054, Email: lbszjzx@bmlp.com |


### Example 3 — 4 turns on one image

![case3](samples/case3.png)

| # | question | answer |
|---|---|---|
| 1 | \nAnalyze this document and identify the structured information and relationships. | I can identify the following key information from this document:\n\n**Key-Value Relationships:**\n• 报告单号:: 2017102034\n• 产品名称: 13X分子筛\n• 生产日期: 2017-10-20\n• 执行标准: HG/T2690-2012\n• 抽样日期: 2017-10-20\n• 产品批量: 100吨\n• 化验日期: 2017-10-20\n• 生产单位:: 巩义市豫润海源净水材料有限公司\n• 型号: 13X分子筛\n• 型号: 13X分子筛 |
| 2 | What are the specific test results and standards for Static Water Adsorption, Packaging Moisture Content, Carbon Dioxide Adsorption Capacity, and Abrasion Rate? | Static Water Adsorption: Standard ≥25%, Result 30%. Packaging Moisture Content: Standard ≤1.5%, Result 1.2%. Carbon Dioxide Adsorption Capacity: Standard ≥18, Result 20%. Abrasion Rate: Standard ≤0.2%, Result 0.18%. |
| 3 | What is the conclusion of the inspection report and what does the red stamp say? | The conclusion is '产品各项合格' (Product items qualified), and the red stamp reads '检验合格' (Inspection Qualified). |
| 4 | What is the English name of the company listed in the header? | Gongyi city embellish sea water purification materials co., LTD |


### Example 4 — 6 turns on one image

![case4](samples/case4.png)

| # | question | answer (cleaned) | org_answer (pre-judge) |
|---|---|---|---|
| 1 | \nExtract all key-value pairs and important information from this document. | **Key-Value Relationships:**<br>• 座位号: 23A<br>• 序号: 023<br>• 航班: CA 4498<br>• 舱位: S<br>• 日期: 11JUL<br>• 登机时间: 0900<br>• 登机口: 4<br>• 始发地: JZH 九寨黄龙<br>• 目的地: CTU 成都<br>• 姓名: YANGYI 杨亦<br>• 身份识别: NI3306021999...102X<br>• 票号: 9995283376205/1<br>• 机场: 四川九寨黄龙机场 (Sichuan jiuzhai huanglong airport) | I can identify the following key information from this document:\n\n**Key-Value Relationships:**\n• 座位号: 23A\n• 序号: 023\n• 航班: CA 4498\n• 舱位: S\n• 日期: 11JUL\n• 登机时间: 0900\n• 登机口: 4\n• 始发地: JCH 九寨黄龙\n• 目的地: CTU 成都\n• 身份识别: 102X |
| 2 | What is the ticket number (TKT NO) listed on the boarding pass? | 9995283376205/1 | _(unchanged)_ |
| 3 | What is the passenger's name? | YANGYI / 杨亦 | _(unchanged)_ |
| 4 | What is the full name of the airport at the top of the pass? | 四川九寨黄龙机场 Sichuan jiuzhai huanglong airport | _(unchanged)_ |
| 5 | What is the warning text at the bottom regarding gate closing? | 登机口于起飞前10分钟关闭 GATES CLOSE 10 MINUTES BEFORE DEPARTURE TIME | _(unchanged)_ |
| 6 | What is the watermark text in the bottom right corner? | 7788收藏@S13334-2020.07.10 | _(unchanged)_ |


### Example 5 — 4 turns on one image

![case5](samples/case5.png)

| # | question | answer (cleaned) | org_answer (pre-judge) |
|---|---|---|---|
| 1 | \nExtract all key-value pairs and important information from this document. | **Key-Value Relationships:**<br>• 座位号: 8C<br>• 舱位: J<br>• 日期: 26MAY<br>• 序号: 012<br>• 航班: JR1632<br>• 始发地: 铜仁 (TONGREN FENGHUANG)<br>• 目的地: 长沙 (CHANGSHA)<br>• 登机时间: 0930<br>• 登机口: 02<br>• ETKT: 9292093909297/1 | I can identify the following key information from this document:\n\n**Key-Value Relationships:**\n• 座位号: 8C\n• 舱位: J\n• 日期: 26MAY\n• 序号: 012\n• 航班: JR1632\n• 始发地: 铜仁\n• 目的地: 长沙\n• 登机时间: 0930\n• 登机口: 02\n• ETKT: 9292093909297/1 |
| 2 | \nExtract all the named entities and classify them by type. | **Entities Identified:**<br>• Title: 登机牌 BOARDING PASS<br>• Flight Number: JR1632<br>• Date: 26MAY<br>• Class: J<br>• Serial No.: 012<br>• Seat No.: 8C<br>• Destination: 长沙 CHANGSHA<br>• Origin: 铜仁 TONGREN FENGHUANG<br>• Gate: 02<br>• Boarding Time: 0930<br>• Ticket No.: 9292093909297/1 | **Entities Identified:**\n• Title, which describes the type of cards, tickets, bills, papers and so on. It usually locates on the top part of the first row.: PASS, BOARDING, 登机牌 |
| 3 | What are the English names for the destination and origin cities listed on the boarding pass? | Destination: CHANGSHA; Origin: TONGREN FENGHUANG. | _(unchanged)_ |
| 4 | What is the warning text printed at the very bottom of the boarding pass? | 登机口于起飞前15分钟关闭 GATES CLOSE 15 MINUTES BEFORE DEPARTURE TIME | _(unchanged)_ |


### Example 6 — 6 turns on one image

![case6](samples/case6.png)

| # | question | answer (cleaned) | org_answer (pre-judge) |
|---|---|---|---|
| 1 | \nWhat are the key details and their values in this document? | I can identify the following key information from this document:<br><br>**Key-Value Relationships:**<br>• 24小时订票热线: 0731-84112222<br>• 航班号: MF8221<br>• 始发站: 长沙<br>• 到达站: 咸阳<br>• ORIGIN: CHANGSHA<br>• DEST: XI AN/XIANYANG<br>• 日期: 10MAR<br>• 登机时间: 1730<br>• 座位号: 15A<br>• 舱位: G<br>• 登机口: 42<br>• ETKT: 7312312246845/1<br>• ID: NI43018119870221504X | I can identify the following key information from this document:\n\n**Key-Value Relationships:**\n• 24小时订票热线:: 0731-84112222\n• 航班号: MF8221\n• 始发站: 长沙\n• 到达站: 咸阳\n• ORIGIN: CHANGSHA\n• DEST: XI AN/XI\n• 日期: 10MAR\n• 登机时间: 1730\n• 座位号: 15A\n• 登机口: 42 |
| 2 | \nAnalyze this document and identify all key entities with their classifications. | **Entities Identified:**<br>• Institution: CHANGSHA HUANGHUA INTERNATIONAL AIRPORT, 长沙黄花国际机场<br>• Flight Number: MF8221<br>• Date: 10MAR<br>• Seat: 15A<br>• Class: G<br>• Gate: 42<br>• ID Number: NI43018119870221504X | **Entities Identified:**\n• Institution, which is the name or appellation of different organization including companies, merchants, business entities, academic community, and so on.: CHANGSHA HUANGHUAINTERNATIONALAIRPORT, 长沙黄花国际机场 |
| 3 | What is the cabin class indicated on the boarding pass? | G | _(unchanged)_ |
| 4 | What is the ETKT number printed at the bottom left? | 7312312246845/1 | _(unchanged)_ |
| 5 | What is the ID number located below the gate number? | NI43018119870221504X | _(unchanged)_ |
| 6 | What is the text inside the red rectangular box on the bottom right? | 登机口可能变更，请注意广播或提示信息。 | _(unchanged)_ |


### Example 7 — 6 turns on one image

![case7](samples/case7.png)

| # | question | answer (cleaned) | org_answer (pre-judge) |
|---|---|---|---|
| 1 | \nWhat types of entities can you detect in this document? | **Entities Identified:**<br>• Title: BOARDING PASS, 登机牌<br>• Institution: 四川九寨黄龙机场, Sichuan jiuzhai huanglong airport<br>• Flight: MU 2374<br>• Date: 15JUL<br>• Class: E<br>• Serial No: 071<br>• Seat No: 36L<br>• Destination: XIY 西安咸阳<br>• Origin: JZH 九寨黄龙<br>• Gate: 5<br>• Boarding Time: 1010<br>• Name: WANGYONG / 汪永<br>• Ticket No: 7812483208128/1 | **Entities Identified:**\n• Title, which describes the type of cards, tickets, bills, papers and so on. It usually locates on the top part of the first row.: BOARDING PASS, 登机牌\n• Institution, which is the name or appellation of different organization including companies, merchants, business entities, academic community, and so on.: 四川九寨黄龙机场, Sichuan jiuzhai huanglong aiport |
| 2 | \nExtract all the named entities and classify them by type. | **Entities Identified:**<br>• Title: BOARDING PASS, 登机牌<br>• Institution: 四川九寨黄龙机场, Sichuan jiuzhai huanglong airport<br>• Flight: MU 2374<br>• Date: 15JUL<br>• Class: E<br>• Serial No: 071<br>• Seat No: 36L<br>• Destination: XIY 西安咸阳<br>• Origin: JZH 九寨黄龙<br>• Gate: 5<br>• Boarding Time: 1010<br>• Name: WANGYONG / 汪永<br>• Ticket No: 7812483208128/1 | **Entities Identified:**\n• Title, which describes the type of cards, tickets, bills, papers and so on. It usually locates on the top part of the first row.: BOARDING, BOARDING PASS, 机牌\n• Institution, which is the name or appellation of different organization including companies, merchants, business entities, academic community, and so on.: 四川九寨黄龙机场, Sichuan jiuzhai huanglong ainport |
| 3 | \nPlease extract and organize all the important information from this document. | **Key-Value Relationships:**<br>• 航班 (FLIGHT): MU 2374<br>• 日期 (DATE): 15JUL<br>• 舱位 (CLASS): E<br>• 序号 (SERIAL NO.): 071<br>• 座位号 (SEAT NO.): 36L<br>• 目的地 (TO): XIY 西安咸阳<br>• 始发地 (FROM): JZH 九寨黄龙<br>• 登机口 (GATE): 5<br>• 登机时间 (BDT): 1010<br>• 姓名 (NAME): WANGYONG / 汪永<br>• 票号 (TKT NO.): 7812483208128/1 | I can identify the following key information from this document:\n\n**Key-Value Relationships:**\n• 航班: MU 2374\n• 航班: MU\n• 日期: 15JUL\n• 舱位: E\n• 序号: 071\n• 座位号: 36L\n• 日期: JIL\n• 目的地: 西安咸阳\n• 始发地: JZH九寨黄龙\n• 登机时间: 1010 |
| 4 | What is the gate number and the passenger's name listed on the boarding pass? | Gate: 5, Name: WANGYONG / 汪永 | _(unchanged)_ |
| 5 | What is the ticket number (TKT NO) and the warning text at the bottom of the pass? | TKT NO: 7812483208128/1, Warning: GATES CLOSE 10 MINUTES BEFORE DEPARTURE TIME | _(unchanged)_ |
| 6 | What website watermark is visible in the top right corner of the image? | 7788.com | _(unchanged)_ |


### Example 8 — 8 turns on one image

![case8](samples/case8.png)

| # | question | answer (cleaned) | org_answer (pre-judge) |
|---|---|---|---|
| 1 | \nWhat types of entities can you detect in this document? | **Entities Identified:**\n• Title, which describes the type of cards, tickets, bills, papers and so on. It usually locates on the top part of the first row.: 2015年出口日本茶叶推荐使用农药清单 | _(unchanged)_ |
| 2 | \nExtract all key-value pairs and important information from this document. | The document contains a table with 25 rows of pesticide data. Key information includes:<br>1. Bifenthrin (联苯菊酯), Limit: 30, Target: 防治茶黑毒蛾、象甲类、茶尺蠖、茶蝉类害虫<br>2. Lambda-cypermethrin (高效氯氰菊酯), Limit: 20, Target: 防治茶尺蠖、茶毛虫、茶细蛾等鳞翅目害虫，假眼小绿叶蝉、茶叶螨类害虫<br>3. Lambda-Cyhalothrin (高效氯氟氰菊酯、功夫菊酯), Limit: 15, Target: 防治茶尺蠖、茶毛虫、茶细蛾等鳞翅目害虫，假眼小绿叶蝉、茶叶螨类害虫<br>4. Thiacloprid (噻虫啉), Limit: 30, Target: 茶小绿叶蝉、黑粉刺虱等害虫<br>5. Cypermethrin (氯氰菊酯、安绿宝、高宝), Limit: 20, Target: 防治茶尺蠖、茶毛虫、刺蛾类、卷叶蛾类害虫<br>6. Dinotefuran (呋虫胺), Limit: 25, Target: 茶小绿叶蝉<br>7. Flufenoxuron (氟虫脲), Limit: 15, Target: 防治茶小绿叶蝉、茶尺蠖、茶毛虫、茶细蛾<br>8. Buprofezin (噻嗪酮、优乐得、扑虱灵), Limit: 30, Target: 防治叶蝉、蓟马、粉虱、蚧类，兼治茶橙瘿螨类<br>9. Diafenthiuron (丁醚脲), Limit: 20, Target: 茶小绿叶蝉、茶尺蠖等鳞翅目害虫<br>10. Tebufenozide+chlorfenapyr (虫酰肼+虫螨腈、黄药师), Limit: 25+40, Target: 茶小绿叶蝉、茶尺蠖等鳞翅目害虫<br>11. Flonicamid (氟啶虫酰胺), Limit: 40, Target: 茶小绿叶蝉、蓟马、红蜘蛛<br>12. Propargite (克螨特、炔螨特), Limit: 5, Target: 防治茶螨类害虫<br>13. Imidacloprid (吡虫啉、一遍净、大功臣、蚜虱净), Limit: 10, Target: 防治茶蚜、黑粉刺虱、假眼小绿叶蝉、茶叶蛾类害虫<br>14. Pyridaben (哒螨灵、扫螨净), Limit: 10, Target: 防治茶螨类害虫<br>15. Fenpropathrin (甲氰菊酯、灭扫利), Limit: 25, Target: 防治鳞翅目害虫、茶蚜、茶蚧、粉虱类害虫<br>16. Deltamethrin (溴氰菊酯), Limit: 10, Target: 防治毒蛾类、尺蠖、茶蚜、黑刺粉虱类、蚧类害虫<br>17. Tridemorph (十三吗啉), Limit: 20, Target: 炭疽病、茶饼病、白星病、茶云纹叶枯病等叶面病害<br>18. Chlorothalonil (百菌清、达科宁), Limit: 10, Target: 防治炭疽病、茶白星病、茶云纹叶枯病、茶叶斑病<br>19. Thiamethoxam (噻虫嗪), Limit: 20, Target: 茶小绿叶蝉、黑粉刺虱<br>20. Chlorfenapyr (溴虫腈、虫螨腈), Limit: 40, Target: 防治茶小绿叶蝉、茶叶螨类害虫<br>21. Becillus thuringiensis (苏云金杆菌（B·T）), Limit: 豁免物质, Target: 防治茶毛虫、茶黑毒蛾、茶刺蛾等鳞翅目幼虫<br>22. Bordeaux mixture (波尔多液), Limit: 豁免物质, Target: 封园防病虫害（每两年使用一次）<br>23. Calium polysulfides (石硫合剂), Limit: 豁免物质, Target: 封园防病虫害（每两年使用一次）<br>24. Petroleum oil (绿颖), Limit: 豁免物质, Target: 红蜘蛛、青苔、煤烟病<br>25. Copper hydroxide (氢氧化铜), Limit: 豁免物质, Target: 苔藓、地衣 | I can identify the following key information from this document:\n\n**Key-Value Relationships:**\n• 1: 联苯菊酯\n• 1: 30\n• 1: 防治茶黑毒蛾、象甲类、茶尺蠖、茶蝉类害虫\n• 2: 高效氯氰菊酯\n• 2: 20\n• 2: 防治茶尺蠖、茶毛虫、茶细蛾等鳞翅目害虫，假眼小绿叶蝉、茶叶螨类害虫\n• 3: 高效氯氟氰菊酯、功夫菊酯\n• 3: 15\n• 3: 防治茶尺蠖、茶毛虫、茶细蛾等鳞翅目害虫，假眼小绿叶蝉、茶叶螨类害虫\n• 4: 噻虫啉 |
| 3 | What are the column headers of the table? | The column headers are: 序号 (No.), 农药英文名 (Pesticide English Name), 农药中文名 (Pesticide Chinese Name), 日本残留限量标准 (毫克/千克) (Japan Residue Limit Standard (mg/kg)), and 防治对象 (Target Pests/Diseases). | _(unchanged)_ |
| 4 | List the pesticide details for rows 5 through 10. | 5. Cypermethrin (氯氰菊酯、安绿宝、高宝), Limit: 20, Target: 防治茶尺蠖、茶毛虫、刺蛾类、卷叶蛾类害虫<br>6. Dinotefuran (呋虫胺), Limit: 25, Target: 茶小绿叶蝉<br>7. Flufenoxuron (氟虫脲), Limit: 15, Target: 防治茶小绿叶蝉、茶尺蠖、茶毛虫、茶细蛾<br>8. Buprofezin (噻嗪酮、优乐得、扑虱灵), Limit: 30, Target: 防治叶蝉、蓟马、粉虱、蚧类，兼治茶橙瘿螨类<br>9. Diafenthiuron (丁醚脲), Limit: 20, Target: 茶小绿叶蝉、茶尺蠖等鳞翅目害虫<br>10. Tebufenozide+chlorfenapyr (虫酰肼+虫螨腈、黄药师), Limit: 25+40, Target: 茶小绿叶蝉、茶尺蠖等鳞翅目害虫 | _(unchanged)_ |
| 5 | List the pesticide details for rows 11 through 20. | 11. Flonicamid (氟啶虫酰胺), Limit: 40, Target: 茶小绿叶蝉、蓟马、红蜘蛛<br>12. Propargite (克螨特、炔螨特), Limit: 5, Target: 防治茶螨类害虫<br>13. Imidacloprid (吡虫啉、一遍净、大功臣、蚜虱净), Limit: 10, Target: 防治茶蚜、黑粉刺虱、假眼小绿叶蝉、茶叶蛾类害虫<br>14. Pyridaben (哒螨灵、扫螨净), Limit: 10, Target: 防治茶螨类害虫<br>15. Fenpropathrin (甲氰菊酯、灭扫利), Limit: 25, Target: 防治鳞翅目害虫、茶蚜、茶蚧、粉虱类害虫<br>16. Deltamethrin (溴氰菊酯), Limit: 10, Target: 防治毒蛾类、尺蠖、茶蚜、黑刺粉虱类、蚧类害虫<br>17. Tridemorph (十三吗啉), Limit: 20, Target: 炭疽病、茶饼病、白星病、茶云纹叶枯病等叶面病害<br>18. Chlorothalonil (百菌清、达科宁), Limit: 10, Target: 防治炭疽病、茶白星病、茶云纹叶枯病、茶叶斑病<br>19. Thiamethoxam (噻虫嗪), Limit: 20, Target: 茶小绿叶蝉、黑粉刺虱<br>20. Chlorfenapyr (溴虫腈、虫螨腈), Limit: 40, Target: 防治茶小绿叶蝉、茶叶螨类害虫 | _(unchanged)_ |
| 6 | List the pesticide details for rows 21 through 25, noting the special category in the limit column. | Rows 21-25 are categorized as '豁免物质' (Exempt substances) in the limit column.<br>21. Becillus thuringiensis (苏云金杆菌（B·T）), Target: 防治茶毛虫、茶黑毒蛾、茶刺蛾等鳞翅目幼虫<br>22. Bordeaux mixture (波尔多液), Target: 封园防病虫害（每两年使用一次）<br>23. Calium polysulfides (石硫合剂), Target: 封园防病虫害（每两年使用一次）<br>24. Petroleum oil (绿颖), Target: 红蜘蛛、青苔、煤烟病<br>25. Copper hydroxide (氢氧化铜), Target: 苔藓、地衣 | _(unchanged)_ |
| 7 | Which rows are highlighted in yellow and which are highlighted in green? | Rows 2, 6, 9, 10, 11, 24, and 25 are highlighted in yellow. Rows 22 and 23 are highlighted in green. | _(unchanged)_ |
| 8 | What is the watermark text visible at the bottom right of the image? | 知乎 @欧阳林三 | _(unchanged)_ |


### Example 9 — 6 turns on one image

![case9](samples/case9.png)

| # | question | answer (cleaned) | org_answer (pre-judge) |
|---|---|---|---|
| 1 | \nExtract all the named entities and classify them by type. | **Entities Identified:**<br>• Title: 装修材料清单预算表<br>• Subtitle: 家庭装修材料清单 预算价格表明细 | **Entities Identified:**\n• Title, which describes the type of cards, tickets, bills, papers and so on. It usually locates on the top part of the first row.: 商都县行政事业性收费公示清单\n• Date, which is the date of the special time node such as departure, arriving, issuing, trading, and so on.: 2017年4月5日 |
| 2 | \nWhat are the key details and their values in this document? | **Key-Value Relationships (Partial Extraction):**<br>• Location: 客<br>• Name: 顶面乳胶漆, Quantity: 22, Unit Price: 12, Amount: 264<br>• Name: 客厅吊顶, Quantity: 9, Unit Price: 130, Amount: 1170<br>• Name: 顶面造型吊顶, Quantity: 30.04, Unit Price: 95, Amount: 2853.8 | I can identify the following key information from this document:\n\n**Key-Value Relationships:**\n• 工程量: 22\n• 单价: 12\n• 金额: 264\n• 名称: 顶面乳胶漆\n• 工程量: 9\n• 单价: 130\n• 金额: 1170\n• 名称: 客厅吊顶\n• 名称: 顶面造型吊顶\n• 工程量: 30.04 |
| 3 | What is the main title of the document? | 装修材料清单预算表 | _(unchanged)_ |
| 4 | What are the column headers in the table? | 位置, 名称, 工程量 (平方米), 单价 (元), 金额 (元), 备注 | _(unchanged)_ |
| 5 | List the items and their total amounts for the '厨' (Kitchen) section. | 橱柜地柜: 2800, 吊柜: 800, 洗菜盆及高弯龙头: 360, 抽油烟机: 1200, 炉盘: 420, 吸顶灯: 100, 热水器: 1000 | _(unchanged)_ |
| 6 | What are the items listed under the '主' (Master Bedroom) location? | 顶面基层处理, 顶面乳胶漆, 墙面乳胶漆 | _(unchanged)_ |


### Example 10 — 7 turns on one image

![case10](samples/case10.png)

| # | question | answer (cleaned) | org_answer (pre-judge) |
|---|---|---|---|
| 1 | \nAnalyze this document and identify the structured information and relationships. | **Key-Value Relationships:**<br>• 招标人及工程名称: 锡林郭勒盟苏尼特左旗2015年公共租赁住房恩格尔小区内配套基础设施建设项目<br>• 招标编号: TQCGGC2016-DS-GKZB-SZ-SG002<br>• 建设地点: 苏尼特左旗满都拉图镇<br>• 资金来源: 申请中央预算内投资<br>• 招标范围: 图纸设计的各工程内容<br>• 计划工期: 2016年9月-2016年12月<br>• 建设规模: 供热管道300米、排水管道300米小区硬化及道路硬化3500㎡<br>• 投标人资质要求: 施工投标人须具备独立的法人资格，具备市政公用工程施工总承包叁级(含暂叁级)及以上资质的企业。(不接受联合体投标)<br>• 报名地点: 网上报名<br>• 招标文件的获取: 1、资格审查合格后方可购买招标文件 2、招标文件每套售价施工1000元，售后不退。<br>• 发布公告媒介: 中国采购与招标网(www.chinabidding.com.cn)、锡林郭勒盟政务服务(公共资源交易)中心网站http://www.xmzwggzy.com、太仆寺旗公共资源交易中心网(http://www.tqzwfw.com/tq/default.aspx)<br>• 招标代理机构名称: 亿诚建设项目管理有限公司<br>• 联系电话: 15249508558 | I can identify the following key information from this document:\n\n**Key-Value Relationships:**\n• 招标人及工程名称: 锡林郭勒盟苏尼特左旗2015年公共租赁住房恩格尔小区内配套基础设施建设\n• 建设地点: 苏尼特左旗满都拉图镇\n• 资金: 申请中央预算内投资\n• 招标范围: 图纸设计的各工程内容\n• 计划: 2016年9月-2016年12月\n• 建设规模: 供热管道300米、排水管道300米小区硬化及道路硬化3500㎡\n• 投标人资质要求: 施工投标人须具各独立的法人资格，具备市政公用工程施工总承包叁级(含\n• 报名地点: 网上报名\n• 招标文件的获取: 1、资格审查合格后方可购买招标文件\n• 发布公告: 中国采购与招标网(www.chinabidding.com.cn)、锡林郭勒盟政务服务 |
| 2 | What is the main title of the document at the top? | 招标公告审核备案表 | _(unchanged)_ |
| 3 | What is the bidding number (招标编号) listed in the first row? | TQCGGC2016-DS-GKZB-SZ-SG002 | _(unchanged)_ |
| 4 | What is the name of the bidding agency (招标代理机构名称)? | 亿诚建设项目管理有限公司 | _(unchanged)_ |
| 5 | What is the contact phone number (联系电话) provided in the table? | 15249508558 | _(unchanged)_ |
| 6 | What is the price for each set of bidding documents (招标文件每套售价)? | 1000元 | _(unchanged)_ |
| 7 | What is the handwritten date visible at the bottom of the form? | 2016年8月26日 | _(unchanged)_ |