[Bug]: After receiving mineru, the error "FAILED TO PARSE TABLE" occurred during the parsing process.

[yuwei88888]

Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (Language Policy).
• Non-english title submitions will be closed directly ( 非英文标题的提交将会被直接关闭 ) (Language Policy).
• Please do not modify this template :) and fill in all the required fields.

RAGFlow workspace code commit ID

none

RAGFlow image version

v0.25.6

Other environment information

Actual behavior

2026-06-17 09:49:42,720 INFO 33 [HISTORY][
{
"role": "system",
"content": "You are a robust Table-of-Contents (TOC) extractor.\n\nGOAL\nGiven a dictionary of chunks
{"<chunk_ID>": chunk_text}, extract TOC-like headings and return a strict JSON array of objects:\n[\n {"title":
"", "chunk_id": ""},\n ...\n]\n\nFIELDS\n- "title": the heading text (clean, no page numbers or leader dots).\n

• If any part of a chunk has no valid heading, output that part as {"title":"-1", ...}.\n- "chunk_id": the chunk
  ID (string).\n - One chunk can yield multiple JSON objects in order (unmatched text + one or more
  headings).\n\nRULES\n1) Preserve input chunk order strictly.\n2) If a chunk contains multiple headings, expand them in
  order:\n - Pre-heading narrative → {"title":"-1","chunk_id":"<chunk_ID>"}\n - Then each heading →
  {"title":"...","chunk_id":"<chunk_ID>"}\n3) Do not merge outputs across chunks; each object refers to exactly one
  chunk ID.\n4) "title" must be non-empty (or exactly "-1"). "chunk_id" must be a string (chunk ID).\n5) When
  ambiguous, prefer "-1" unless the text strongly looks like a heading.\n\nHEADING DETECTION (cues, not hard rules)\n-
  Appears near line start, short isolated phrase, often followed by content.\n- May contain separators: — —— - : ： · •\n-
  Numbering styles:\n • 第[一二三四五六七八九十百]+(篇|章|节|条)\n • [(（]?[一二三四五六七八九十]+[)）]?\n •
  [(（]?[①②③④⑤⑥⑦⑧⑨⑩][)）]?\n • ^\d+(\.\d+)[)．.]?\s\n • ^[IVXLCDM]+[).]\n • ^[A-Z][).]\n- Canonical section cues
  (general only):\n Common heading indicators include words such as:\n "Overview", "Introduction", "Background",
  "Purpose", "Scope", "Definition",\n "Method", "Procedure", "Result", "Discussion", "Summary",
  "Conclusion",\n "Appendix", "Reference", "Annex", "Acknowledgment", "Disclaimer".\n These are soft cues,
  not strict requirements.\n- Length restriction:\n • Chinese heading: ≤25 characters\n • English heading: ≤80
  characters\n- Exclude long narrative sentences, continuous prose, or bullet-style lists → output as "-1".\n\nOUTPUT
  FORMAT\n- Return ONLY a valid JSON array of {"title","content"} objects.\n- No reasoning or
  commentary.\n\nEXAMPLES\n\nExample 1 — No heading\nInput:\n[{"0": "Copyright page · Publication info (ISBN 123-456).
  All rights reserved."}, ...]\nOutput:\n[\n {"title":"-1","chunk_id":"0"},\n ...\n]\n\nExample 2 — One
  heading\nInput:\n[{"1": "Chapter 1: General Provisions This chapter defines the overall rules…"}, ...]\nOutput:\n[\n
  {"title":"Chapter 1: General Provisions","chunk_id":"1"},\n ...\n]\n\nExample 3 — Narrative +
  heading\nInput:\n[{"2": "This paragraph introduces the background and goals. Section 2: Definitions Key terms are
  explained…"}, ...]\nOutput:\n[\n {"title":"Section 2: Definitions","chunk_id":"2"},\n ...\n]\n\nExample 4 —
  Multiple headings in one chunk\nInput:\n[{"3": "Declarations and Commitments (I) Party B commits… (II) Party C
  commits… Appendix A Data Specification"}, ...]\nOutput:\n[\n {"title":"Declarations and
  Commitments","chunk_id":"3"},\n {"title":"(I) Party B commits","chunk_id":"3"},\n {"title":"(II) Party
  C commits","chunk_id":"3"},\n {"title":"Appendix A Data Specification","chunk_id":"3"},\n
  ...\n]\n\nExample 5 — Numbering styles\nInput:\n[{"4": "1. Scope: Defines boundaries. 2) Definitions: Terms used. III)
  Methods Overview."}, ...]\nOutput:\n[\n {"title":"1. Scope","chunk_id":"4"},\n {"title":"2)
  Definitions","chunk_id":"4"},\n {"title":"III) Methods Overview","chunk_id":"4"},\n ...\n]\n\nExample 6 —
  Long list (NOT headings)\nInput:\n{"5": "Item list: apples, bananas, strawberries, blueberries, mangos, peaches"},
  ...]\nOutput:\n[\n {"title":"-1","chunk_id":"5"},\n ...\n]\n\nExample 7 — Mixed
  Chinese/English\nInput:\n{"6": "（出版信息略）This standard follows industry practices. Chapter 1: Overview 摘要…
  第2节：术语与缩略语"}, ...]\nOutput:\n[\n {"title":"Chapter 1: Overview","chunk_id":"6"},\n
  {"title":"第2节：术语与缩略语","chunk_id":"6"},\n ...\n]"
  },
  {
  "role": "user",
  "content": "OUTPUT FORMAT\n- Return ONLY the JSON array.\n- Use double quotes.\n- No extra commentary.\n- Keep
  language of "title" the same as the input.\n\nINPUT\n{"1": "-11收入12650.9013.001435.97销售部178720.85\nCW-20260422
  026-02-12支出4190.809.00341.17采购部174529.05\nCW-20260432026-02-13收入9580.706.00522.56市场部184109.75\nCW-20260442026
  -02-14支出2670.606.00146.08行政部181439.15\n收入15120.8013.001700.56销售部196559.95\nCW- 20260452026-02- 15\nFAILED TO
  PARSE TABLE\nFAILED TO PARSE TABLE\nFAILED TO PARSE TABLE\n补充说明：1.表格共45⾏数据，刚好填满两⻚A4 PDF；2.所有数值
  均为财务常⽤合理数值，间隔清晰，可直接编辑修改；3.适配所有主流PDF编辑软件，下载后即可编辑。\n（注：⽂档部分内容可能由 AI
  ⽣成）"}"
  }

Expected behavior

No response

Steps to reproduce

1.According to the configuration file in Picture One
2.Parse the file and return the success log
09:49:28 Task has been received.
09:49:28 Page(1~5): Start to parse.
09:49:28 Page(1~5): [MinerU] Received binary PDF -> /tmp/mineru_bin_pdf_bq2rmc1a/测试的可编辑表格6.pdf
09:49:29 Page(1~5): [MinerU] Output directory: /tmp/111/out
09:49:30 Page(1~5): [MinerU] invoke api: http://101.36.73.92:8091/file_parse
09:49:37 Page(1~5): [MinerU] zip file returned, saving to /tmp/111/out/测试的可编辑表格6_auto_10nvgyv7.zip...
09:49:37 Page(1~5): [MinerU] Unzip to /tmp/111/out/测试的可编辑表格6_auto_10nvgyv7...
09:49:37 Page(1~5): [MinerU] Parsed 8 blocks from PDF.
09:49:37 Page(1~5): Finish parsing.
09:49:39 Page(1~5): Start to generate keywords for every chunk ...
09:49:40 Page(1~5): Keywords generation 2 chunks completed in 1.46s
09:49:40 Page(1~5): Generate 2 chunks
09:49:41 Page(1~5): Embedding chunks (0.89s)
09:49:42 Page(1~5): Start to generate table of content ...
09:49:43 Page(1~5): Indexing done (0.91s).
09:49:47 Page(1~5): Task done (19.90s)
3.The analysis was successful, but no tabular form was generated, as shown in Figure 2.

Additional information

Picture 1

Picture 2

Overall log
2026-06-17 09:49:28,700 INFO 33 Parsed MinerU config (sensitive fields redacted): {'llm_name': 'MinerU', 'mineru_apiserver': 'http://101.36.73.92:8091', 'mineru_output_dir': '/tmp/111/out', 'mineru_backend': 'pipeline', 'mineru_delete_output': '0'}
2026-06-17 09:49:28,738 INFO 33 [MinerU] API openapi.json reachable=True url=http://101.36.73.92:8091/openapi.json
2026-06-17 09:49:28,740 INFO 33 [MinerU] Received binary PDF -> /tmp/mineru_bin_pdf_bq2rmc1a/测试的可编辑表格6.pdf
2026-06-17 09:49:29,666 INFO 33 set_progress(c53b3c9a69ee11f1b8b85533fa24cb09), progress: 0.15, progress_msg: 09:49:28 Page(15): [MinerU] Received binary PDF -> /tmp/mineru_bin_pdf_bq2rmc1a/测试的可编辑表格6.pdf
2026-06-17 09:49:29,666 INFO 33 [MinerU] Output directory: /tmp/111/out backend=pipeline api=http://101.36.73.92:8091 server_url=
2026-06-17 09:49:30,015 INFO 33 set_progress(c53b3c9a69ee11f1b8b85533fa24cb09), progress: 0.15, progress_msg: 09:49:29 Page(15): [MinerU] Output directory: /tmp/111/out
2026-06-17 09:49:30,120 INFO 33 [MinerU] request data={'output_dir': './output', 'lang_list': <MinerULanguage.CH: 'ch'>, 'backend': <MinerUBackend.PIPELINE: 'pipeline'>, 'parse_method': <MinerUParseMethod.AUTO: 'auto'>, 'formula_enable': False, 'table_enable': True, 'server_url': None, 'return_md': True, 'return_middle_json': True, 'return_model_output': True, 'return_content_list': True, 'return_images': True, 'response_format_zip': True, 'start_page_id': 0, 'end_page_id': 99999}
2026-06-17 09:49:30,120 INFO 33 [MinerU] request options=MinerUParseOptions(backend=<MinerUBackend.PIPELINE: 'pipeline'>, lang=<MinerULanguage.CH: 'ch'>, method=<MinerUParseMethod.AUTO: 'auto'>, server_url='', delete_output=False, parse_method='raw', formula_enable=False, table_enable=True)
2026-06-17 09:49:30,120 INFO 33 [MinerU] invoke api: http://101.36.73.92:8091/file_parse backend=pipeline server_url=None
2026-06-17 09:49:30,299 INFO 33 set_progress(c53b3c9a69ee11f1b8b85533fa24cb09), progress: 0.2, progress_msg: 09:49:30 Page(15): [MinerU] invoke api: http://101.36.73.92:8091/file_parse
2026-06-17 09:49:31,245 INFO 28 HEAD http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0 [status:200 duration:0.002s]
2026-06-17 09:49:31,252 INFO 28 POST http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0/_search [status:200 duration:0.004s]
[2026-06-17 09:49:31 +0800] [28] [INFO] 127.0.0.1:40506 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6/documents 1.1 200 26772 22363
2026-06-17 09:49:31,269 INFO 28 HEAD http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0 [status:200 duration:0.001s]
2026-06-17 09:49:31,275 INFO 28 POST http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0/_search [status:200 duration:0.003s]
[2026-06-17 09:49:31 +0800] [28] [INFO] 127.0.0.1:40516 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6/documents 1.1 200 22553 18364
[2026-06-17 09:49:31 +0800] [28] [INFO] 127.0.0.1:40522 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6 1.1 200 2360 10523
2026-06-17 09:49:36,280 INFO 28 HEAD http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0 [status:200 duration:0.002s]
2026-06-17 09:49:36,286 INFO 28 POST http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0/_search [status:200 duration:0.004s]
[2026-06-17 09:49:36 +0800] [28] [INFO] 127.0.0.1:40536 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6/documents 1.1 200 27047 22392
2026-06-17 09:49:36,304 INFO 28 HEAD http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0 [status:200 duration:0.002s]
2026-06-17 09:49:36,309 INFO 28 POST http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0/_search [status:200 duration:0.003s]
[2026-06-17 09:49:36 +0800] [28] [INFO] 127.0.0.1:40542 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6/documents 1.1 200 22553 18041
[2026-06-17 09:49:36 +0800] [28] [INFO] 127.0.0.1:40550 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6 1.1 200 2360 9901
2026-06-17 09:49:37,144 INFO 33 [MinerU] zip file returned, saving to /tmp/111/out/测试的可编辑表格6_auto_10nvgyv7.zip...
2026-06-17 09:49:37,432 INFO 33 set_progress(c53b3c9a69ee11f1b8b85533fa24cb09), progress: 0.3, progress_msg: 09:49:37 Page(15): [MinerU] zip file returned, saving to /tmp/111/out/测试的可编辑表格6_auto_10nvgyv7.zip...
2026-06-17 09:49:37,534 INFO 33 [MinerU] Unzip to /tmp/111/out/测试的可编辑表格6_auto_10nvgyv7...
2026-06-17 09:49:37,534 INFO 33 [MinerU] Extract zip: zip_path=/tmp/111/out/测试的可编辑表格6_auto_10nvgyv7.zip, extract_to=/tmp/111/out/测试的可编辑表格6_auto_10nvgyv7, root_hint=测试的可编辑表格6/
2026-06-17 09:49:37,721 INFO 33 set_progress(c53b3c9a69ee11f1b8b85533fa24cb09), progress: 0.4, progress_msg: 09:49:37 Page(15): [MinerU] Unzip to /tmp/111/out/测试的可编辑表格6_auto_10nvgyv7...
2026-06-17 09:49:37,721 INFO 33 [MinerU] Api completed successfully.
2026-06-17 09:49:37,722 INFO 33 [MinerU] Expected output files: 测试的可编辑表格6_content_list.json
2026-06-17 09:49:37,722 INFO 33 [MinerU] Searching output in: /tmp/111/out/测试的可编辑表格6_auto_10nvgyv7
2026-06-17 09:49:37,722 INFO 33 [MinerU] Trying original path: /tmp/111/out/测试的可编辑表格6_auto_10nvgyv7/测试的可编辑表格6_content_list.json
2026-06-17 09:49:37,722 INFO 33 [MinerU] Trying sanitized filename: /tmp/111/out/测试的可编辑表格6_auto_10nvgyv7/测试的可编辑表格6_content_list.json
2026-06-17 09:49:37,722 INFO 33 [MinerU] Trying sanitized nested path: /tmp/111/out/测试的可编辑表格6_auto_10nvgyv7/测试的可编辑表格6/测试的可编辑表格6_content_list.json
2026-06-17 09:49:37,722 INFO 33 [MinerU] Trying vlm subdirectory: /tmp/111/out/测试的可编辑表格6_auto_10nvgyv7/vlm/测试的可编辑表格6_content_list.json
2026-06-17 09:49:37,722 INFO 33 [MinerU] Trying vlm subdirectory with sanitized name: /tmp/111/out/测试的可编辑表格6_auto_10nvgyv7/vlm/测试的可编辑表格6_content_list.json
2026-06-17 09:49:37,722 INFO 33 [MinerU] Trying parse-method path: /tmp/111/out/测试的可编辑表格6_auto_10nvgyv7/auto/测试的可编辑表格6_content_list.json
2026-06-17 09:49:37,723 INFO 33 [MinerU] Parsed 8 blocks from PDF.
2026-06-17 09:49:37,927 INFO 33 set_progress(c53b3c9a69ee11f1b8b85533fa24cb09), progress: 0.75, progress_msg: 09:49:37 Page(15): [MinerU] Parsed 8 blocks from PDF.
2026-06-17 09:49:39,110 INFO 33 set_progress(c53b3c9a69ee11f1b8b85533fa24cb09), progress: 0.8, progress_msg: 09:49:37 Page(15): Finish parsing.
2026-06-17 09:49:39,143 INFO 33 naive_merge(测试的可编辑表格6.pdf): 0.032985937999910675
2026-06-17 09:49:39,143 INFO 33 Chunking(10.850661832002515) 测试的可编辑表格6.pdf/测试的可编辑表格6.pdf done
2026-06-17 09:49:39,329 INFO 33 MINIO PUT(测试的可编辑表格6.pdf) cost 0.186 s
2026-06-17 09:49:39,499 INFO 33 set_progress(c53b3c9a69ee11f1b8b85533fa24cb09), progress: None, progress_msg: 09:49:39 Page(15): Start to generate keywords for every chunk ...
2026-06-17 09:49:39,509 INFO 33 [HISTORY][
{
"role": "system",
"content": "## Role\nYou are a text analyzer.\n\n## Task\nExtract the most important keywords/phrases of a given piece of text content.\n\n## Requirements\n- Summarize the text content, and give the top 3 important keywords/phrases.\n- The keywords MUST be in the same language as the given piece of text content.\n- The keywords are delimited by ENGLISH COMMA.\n- Output keywords ONLY.\n\n---\n\n## Text Content\n\n财务收⽀明细总表 （可编辑）\n说明：本表格为可编辑PDF格式，可直接在WPS、Adobe Acrobat等软件中修改单元格内容，数据涵盖两⻚A4篇幅，均为真实财务类数值，字段清晰、数据间隔合理。\n凭证编号记账日期收支类型业务金额（元）增值税率(%）税额 (元)归属部门累计余额（元）\nCw-20260012026-01-01收入12860.5013.001452.32销售部12860.50\ncw-20260022026-01-02支出3250.759.00267.37行政部9609.75\ncW-20260032026-01-03收入8950.0013.001053.98销售部18559.75\nCW-20260042026-01-04支出1890.206.00105.83财务部16669.55\nCW-20260052026-01-05收入15680.3013.001767.47销售部32349.85\nCW-20260062026-01-06支出4520.809.00371.51采购部27829.05\nCW-20260072026-01-07收入7890.606.00442.50市场部35719.65\ncW-20260082026-01-08支出2150.306.00117.28行政部33569.35\ncW-20260092026-01-09收入11250.8013.001285.42销售部44820.15\ncW-20260102026-01-10支出3860.509.00313.88采购部40959.65\nCW-20260112026-01-11收入9630.206.00536.46市场部50589.85\nCW-20260122026-01-12支出1580.706.0086.48财务部49009.15\n\nCW-20260132026-01-13收入14520.7013.001653.76销售部63529.85\nCW-20260142026-01-14支出5280.309.00431.72采购部58249.55\nCW-20260152026-01-15收入8250.506.00451.69市场部66500.05\nCW-20260162026-01-16支出2760.806.00151.38行政部63739.25\nCW-20260172026-01-17收入13680.9013.001572.76销售部77420.15\nCW-20260182026-01-18支出4120.609.00337.59采购部73299.55\nCW-20260192026-01-19收入10560.406.00581.31市场部83859.95\nCW-20260202026-01-20支出1950.406.00106.22财务部81909.55\nCW-20260212026-01-21收入12350.7013.001406.83销售部94260.25\nCW-20260222026-01-22支出3580.909.00289.88行政部90679.35\nCW-20260232026-01-23收入9870.306.00548.19市场部100549.65\nCW-20260242026-01-24支出2360.506.00128.37财务部98189.15\nCW-20260252026-01-25收入15280.6013.001727.54销售部113469.75\nCW-20260262026-01-26支出4890.709.00397.20采购部108579.05\nCW-20260272026-01-27收入8650.806.00470.15市场部117229.85\nCW-20260282026-01-28支出2580.306.00140.54行政部114649.55\n收入11980.5013.001367.70销售部126630.05\nCW-20260292026-01-29\nCW-20260302026-01-30支出3960.609.00321.35采购部122669.45\nCW-20260312026-02-01收入10250.706.00564.26市场部132920.15\nCW-20260322026-02-02支出2120.406.00115.47财务部130799.75\nCW-20260332026-02-03收入14860.9013.001683.16销售部145660.65\nCW-20260342026-02-04支出4350.809.00352.87采购部141309.85\nCW-20260352026-02-05收入9320.506.00508.35市场部150630.35\nCW-20260362026-02-06支出2890.606.00157.89行政部147739.75\nCW-20260372026-02-07收入13590.8013.001546.82销售部161330.55\nCW-20260382026-02-08支出3680.709.00297.96行政部157649.85\nCW-20260392026-02-09收入10870.606.00597.40市场部168520.45\nCW-20260402026-02-10支出2450.506.00132.33财务部166069.95\nCW-20260412026-02-11收入12650.9013.001435.97销售部178720.85\nCW-20260422026-02-12支出4190.809.00341.17采购部174529.05\nCW-20260432026-02-13收入9580.706.00522.56市场部184109.75\nCW-20260442026-02-14支出2670.606.00146.08行政部181439.15\n收入15120.8013.001700.56销售部196559.95\nCW- 20260452026-02- 15"
},
{
"role": "user",
"content": "Output: "
}
]
09:49:39 - LiteLLM:INFO: utils.py:3995 -
LiteLLM completion() model= qwen3-235b-a22b; provider = dashscope
2026-06-17 09:49:39,533 INFO 33
LiteLLM completion() model= qwen3-235b-a22b; provider = dashscope
2026-06-17 09:49:39,534 INFO 33 [HISTORY][
{
"role": "system",
"content": "## Role\nYou are a text analyzer.\n\n## Task\nExtract the most important keywords/phrases of a given piece of text content.\n\n## Requirements\n- Summarize the text content, and give the top 3 important keywords/phrases.\n- The keywords MUST be in the same language as the given piece of text content.\n- The keywords are delimited by ENGLISH COMMA.\n- Output keywords ONLY.\n\n---\n\n## Text Content\n-11收入12650.9013.001435.97销售部178720.85\nCW-20260422026-02-12支出4190.809.00341.17采购部174529.05\nCW-20260432026-02-13收入9580.706.00522.56市场部184109.75\nCW-20260442026-02-14支出2670.606.00146.08行政部181439.15\n收入15120.8013.001700.56销售部196559.95\nCW- 20260452026-02- 15\nFAILED TO PARSE TABLE\nFAILED TO PARSE TABLE\nFAILED TO PARSE TABLE\n补充说明：1.表格共45⾏数据，刚好填满两⻚A4 PDF；2.所有数值均为财务常⽤合理数值，间隔清晰，可直接编辑修改；3.适配所有主流PDF编辑软件，下载后即可编辑。\n（注：⽂档部分内容可能由 AI ⽣成）"
},
{
"role": "user",
"content": "Output: "
}
]
09:49:39 - LiteLLM:INFO: utils.py:3995 -
LiteLLM completion() model= qwen3-235b-a22b; provider = dashscope
2026-06-17 09:49:39,535 INFO 33
LiteLLM completion() model= qwen3-235b-a22b; provider = dashscope
2026-06-17 09:49:40,929 INFO 33 set_progress(c53b3c9a69ee11f1b8b85533fa24cb09), progress: None, progress_msg: 09:49:40 Page(15): Keywords generation 2 chunks completed in 1.46s
2026-06-17 09:49:40,929 INFO 33 Build document 测试的可编辑表格6.pdf: 12.64s
2026-06-17 09:49:41,099 INFO 33 set_progress(c53b3c9a69ee11f1b8b85533fa24cb09), progress: None, progress_msg: 09:49:40 Page(15): Generate 2 chunks
2026-06-17 09:49:41,316 INFO 28 HEAD http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0 [status:200 duration:0.002s]
2026-06-17 09:49:41,322 INFO 28 POST http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0/_search [status:200 duration:0.003s]
[2026-06-17 09:49:41 +0800] [28] [INFO] 127.0.0.1:47866 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6/documents 1.1 200 27539 21572
2026-06-17 09:49:41,341 INFO 28 HEAD http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0 [status:200 duration:0.001s]
2026-06-17 09:49:41,346 INFO 28 POST http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0/_search [status:200 duration:0.003s]
[2026-06-17 09:49:41 +0800] [28] [INFO] 127.0.0.1:47878 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6/documents 1.1 200 22553 18827
[2026-06-17 09:49:41 +0800] [28] [INFO] 127.0.0.1:47880 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6 1.1 200 2360 11349
2026-06-17 09:49:41,985 INFO 33 set_progress(c53b3c9a69ee11f1b8b85533fa24cb09), progress: 0.7999999999999999, progress_msg:
2026-06-17 09:49:41,985 INFO 33 Embedding chunks (0.89s)
2026-06-17 09:49:42,106 INFO 33 set_progress(c53b3c9a69ee11f1b8b85533fa24cb09), progress: None, progress_msg: 09:49:41 Page(15): Embedding chunks (0.89s)
2026-06-17 09:49:42,261 INFO 33 set_progress(c53b3c9a69ee11f1b8b85533fa24cb09), progress: None, progress_msg: 09:49:42 Page(15): Start to generate table of content ...
2026-06-17 09:49:42,631 INFO 33 set_progress(c53b3c9a69ee11f1b8b85533fa24cb09), progress: None, progress_msg:
2026-06-17 09:49:42,636 INFO 33 [HISTORY][
{
"role": "system",
"content": "You are a robust Table-of-Contents (TOC) extractor.\n\nGOAL\nGiven a dictionary of chunks {"<chunk_ID>": chunk_text}, extract TOC-like headings and return a strict JSON array of objects:\n[\n {"title": "", "chunk_id": ""},\n ...\n]\n\nFIELDS\n- "title": the heading text (clean, no page numbers or leader dots).\n - If any part of a chunk has no valid heading, output that part as {"title":"-1", ...}.\n- "chunk_id": the chunk ID (string).\n - One chunk can yield multiple JSON objects in order (unmatched text + one or more headings).\n\nRULES\n1) Preserve input chunk order strictly.\n2) If a chunk contains multiple headings, expand them in order:\n - Pre-heading narrative → {"title":"-1","chunk_id":"<chunk_ID>"}\n - Then each heading → {"title":"...","chunk_id":"<chunk_ID>"}\n3) Do not merge outputs across chunks; each object refers to exactly one chunk ID.\n4) "title" must be non-empty (or exactly "-1"). "chunk_id" must be a string (chunk ID).\n5) When ambiguous, prefer "-1" unless the text strongly looks like a heading.\n\nHEADING DETECTION (cues, not hard rules)\n- Appears near line start, short isolated phrase, often followed by content.\n- May contain separators: — —— - : ： · •\n- Numbering styles:\n • 第[一二三四五六七八九十百]+(篇|章|节|条)\n • [(（]?[一二三四五六七八九十]+[)）]?\n • [(（]?[①②③④⑤⑥⑦⑧⑨⑩][)）]?\n • ^\d+(\.\d+)[)．.]?\s\n • ^[IVXLCDM]+[).]\n • ^[A-Z][).]\n- Canonical section cues (general only):\n Common heading indicators include words such as:\n "Overview", "Introduction", "Background", "Purpose", "Scope", "Definition",\n "Method", "Procedure", "Result", "Discussion", "Summary", "Conclusion",\n "Appendix", "Reference", "Annex", "Acknowledgment", "Disclaimer".\n These are soft cues, not strict requirements.\n- Length restriction:\n • Chinese heading: ≤25 characters\n • English heading: ≤80 characters\n- Exclude long narrative sentences, continuous prose, or bullet-style lists → output as "-1".\n\nOUTPUT FORMAT\n- Return ONLY a valid JSON array of {"title","content"} objects.\n- No reasoning or commentary.\n\nEXAMPLES\n\nExample 1 — No heading\nInput:\n[{"0": "Copyright page · Publication info (ISBN 123-456). All rights reserved."}, ...]\nOutput:\n[\n {"title":"-1","chunk_id":"0"},\n ...\n]\n\nExample 2 — One heading\nInput:\n[{"1": "Chapter 1: General Provisions This chapter defines the overall rules…"}, ...]\nOutput:\n[\n {"title":"Chapter 1: General Provisions","chunk_id":"1"},\n ...\n]\n\nExample 3 — Narrative + heading\nInput:\n[{"2": "This paragraph introduces the background and goals. Section 2: Definitions Key terms are explained…"}, ...]\nOutput:\n[\n {"title":"Section 2: Definitions","chunk_id":"2"},\n ...\n]\n\nExample 4 — Multiple headings in one chunk\nInput:\n[{"3": "Declarations and Commitments (I) Party B commits… (II) Party C commits… Appendix A Data Specification"}, ...]\nOutput:\n[\n {"title":"Declarations and Commitments","chunk_id":"3"},\n {"title":"(I) Party B commits","chunk_id":"3"},\n {"title":"(II) Party C commits","chunk_id":"3"},\n {"title":"Appendix A Data Specification","chunk_id":"3"},\n ...\n]\n\nExample 5 — Numbering styles\nInput:\n[{"4": "1. Scope: Defines boundaries. 2) Definitions: Terms used. III) Methods Overview."}, ...]\nOutput:\n[\n {"title":"1. Scope","chunk_id":"4"},\n {"title":"2) Definitions","chunk_id":"4"},\n {"title":"III) Methods Overview","chunk_id":"4"},\n ...\n]\n\nExample 6 — Long list (NOT headings)\nInput:\n{"5": "Item list: apples, bananas, strawberries, blueberries, mangos, peaches"}, ...]\nOutput:\n[\n {"title":"-1","chunk_id":"5"},\n ...\n]\n\nExample 7 — Mixed Chinese/English\nInput:\n{"6": "（出版信息略）This standard follows industry practices. Chapter 1: Overview 摘要… 第2节：术语与缩略语"}, ...]\nOutput:\n[\n {"title":"Chapter 1: Overview","chunk_id":"6"},\n {"title":"第2节：术语与缩略语","chunk_id":"6"},\n ...\n]"
},
{
"role": "user",
"content": "OUTPUT FORMAT\n- Return ONLY the JSON array.\n- Use double quotes.\n- No extra commentary.\n- Keep language of "title" the same as the input.\n\nINPUT\n{"0": "\n财务收⽀明细总表 （可编辑）\n说明：本表格为可编辑PDF格式，可直接在WPS、Adobe Acrobat等软件中修改单元格内容，数据涵盖两⻚A4篇幅，均为真实财务类数值，字段清晰、数据间隔合理。\n凭证编号记账日期收支类型业务金额（元）增值税率(%）税额 (元)归属部门累计余额（元）\nCw-20260012026-01-01收入12860.5013.001452.32销售部12860.50\ncw-20260022026-01-02支出3250.759.00267.37行政部9609.75\ncW-20260032026-01-03收入8950.0013.001053.98销售部18559.75\nCW-20260042026-01-04支出1890.206.00105.83财务部16669.55\nCW-20260052026-01-05收入15680.3013.001767.47销售部32349.85\nCW-20260062026-01-06支出4520.809.00371.51采购部27829.05\nCW-20260072026-01-07收入7890.606.00442.50市场部35719.65\ncW-20260082026-01-08支出2150.306.00117.28行政部33569.35\ncW-20260092026-01-09收入11250.8013.001285.42销售部44820.15\ncW-20260102026-01-10支出3860.509.00313.88采购部40959.65\nCW-20260112026-01-11收入9630.206.00536.46市场部50589.85\nCW-20260122026-01-12支出1580.706.0086.48财务部49009.15\n\nCW-20260132026-01-13收入14520.7013.001653.76销售部63529.85\nCW-20260142026-01-14支出5280.309.00431.72采购部58249.55\nCW-20260152026-01-15收入8250.506.00451.69市场部66500.05\nCW-20260162026-01-16支出2760.806.00151.38行政部63739.25\nCW-20260172026-01-17收入13680.9013.001572.76销售部77420.15\nCW-20260182026-01-18支出4120.609.00337.59采购部73299.55\nCW-20260192026-01-19收入10560.406.00581.31市场部83859.95\nCW-20260202026-01-20支出1950.406.00106.22财务部81909.55\nCW-20260212026-01-21收入12350.7013.001406.83销售部94260.25\nCW-20260222026-01-22支出3580.909.00289.88行政部90679.35\nCW-20260232026-01-23收入9870.306.00548.19市场部100549.65\nCW-20260242026-01-24支出2360.506.00128.37财务部98189.15\nCW-20260252026-01-25收入15280.6013.001727.54销售部113469.75\nCW-20260262026-01-26支出4890.709.00397.20采购部108579.05\nCW-20260272026-01-27收入8650.806.00470.15市场部117229.85\nCW-20260282026-01-28支出2580.306.00140.54行政部114649.55\n收入11980.5013.001367.70销售部126630.05\nCW-20260292026-01-29\nCW-20260302026-01-30支出3960.609.00321.35采购部122669.45\nCW-20260312026-02-01收入10250.706.00564.26市场部132920.15\nCW-20260322026-02-02支出2120.406.00115.47财务部130799.75\nCW-20260332026-02-03收入14860.9013.001683.16销售部145660.65\nCW-20260342026-02-04支出4350.809.00352.87采购部141309.85\nCW-20260352026-02-05收入9320.506.00508.35市场部150630.35\nCW-20260362026-02-06支出2890.606.00157.89行政部147739.75\nCW-20260372026-02-07收入13590.8013.001546.82销售部161330.55\nCW-20260382026-02-08支出3680.709.00297.96行政部157649.85\nCW-20260392026-02-09收入10870.606.00597.40市场部168520.45\nCW-20260402026-02-10支出2450.506.00132.33财务部166069.95\nCW-20260412026-02-11收入12650.9013.001435.97销售部178720.85\nCW-20260422026-02-12支出4190.809.00341.17采购部174529.05\nCW-20260432026-02-13收入9580.706.00522.56市场部184109.75\nCW-20260442026-02-14支出2670.606.00146.08行政部181439.15\n收入15120.8013.001700.56销售部196559.95\nCW- 20260452026-02- 15"}"
}
]
09:49:42 - LiteLLM:INFO: utils.py:3995 -
LiteLLM completion() model= qwen3-235b-a22b; provider = dashscope
2026-06-17 09:49:42,637 INFO 33
LiteLLM completion() model= qwen3-235b-a22b; provider = dashscope
2026-06-17 09:49:42,651 INFO 33 PUT http://192.168.80.139:1200/ragflow_105f6de65ce711f182c1edd34d87cce0/_bulk?refresh=wait_for&timeout=60s [status:200 duration:0.542s]
2026-06-17 09:49:42,716 INFO 33 set_progress(c53b3c9a69ee11f1b8b85533fa24cb09), progress: None, progress_msg:
2026-06-17 09:49:42,720 INFO 33 [HISTORY][
{
"role": "system",
"content": "You are a robust Table-of-Contents (TOC) extractor.\n\nGOAL\nGiven a dictionary of chunks {"<chunk_ID>": chunk_text}, extract TOC-like headings and return a strict JSON array of objects:\n[\n {"title": "", "chunk_id": ""},\n ...\n]\n\nFIELDS\n- "title": the heading text (clean, no page numbers or leader dots).\n - If any part of a chunk has no valid heading, output that part as {"title":"-1", ...}.\n- "chunk_id": the chunk ID (string).\n - One chunk can yield multiple JSON objects in order (unmatched text + one or more headings).\n\nRULES\n1) Preserve input chunk order strictly.\n2) If a chunk contains multiple headings, expand them in order:\n - Pre-heading narrative → {"title":"-1","chunk_id":"<chunk_ID>"}\n - Then each heading → {"title":"...","chunk_id":"<chunk_ID>"}\n3) Do not merge outputs across chunks; each object refers to exactly one chunk ID.\n4) "title" must be non-empty (or exactly "-1"). "chunk_id" must be a string (chunk ID).\n5) When ambiguous, prefer "-1" unless the text strongly looks like a heading.\n\nHEADING DETECTION (cues, not hard rules)\n- Appears near line start, short isolated phrase, often followed by content.\n- May contain separators: — —— - : ： · •\n- Numbering styles:\n • 第[一二三四五六七八九十百]+(篇|章|节|条)\n • [(（]?[一二三四五六七八九十]+[)）]?\n • [(（]?[①②③④⑤⑥⑦⑧⑨⑩][)）]?\n • ^\d+(\.\d+)[)．.]?\s\n • ^[IVXLCDM]+[).]\n • ^[A-Z][).]\n- Canonical section cues (general only):\n Common heading indicators include words such as:\n "Overview", "Introduction", "Background", "Purpose", "Scope", "Definition",\n "Method", "Procedure", "Result", "Discussion", "Summary", "Conclusion",\n "Appendix", "Reference", "Annex", "Acknowledgment", "Disclaimer".\n These are soft cues, not strict requirements.\n- Length restriction:\n • Chinese heading: ≤25 characters\n • English heading: ≤80 characters\n- Exclude long narrative sentences, continuous prose, or bullet-style lists → output as "-1".\n\nOUTPUT FORMAT\n- Return ONLY a valid JSON array of {"title","content"} objects.\n- No reasoning or commentary.\n\nEXAMPLES\n\nExample 1 — No heading\nInput:\n[{"0": "Copyright page · Publication info (ISBN 123-456). All rights reserved."}, ...]\nOutput:\n[\n {"title":"-1","chunk_id":"0"},\n ...\n]\n\nExample 2 — One heading\nInput:\n[{"1": "Chapter 1: General Provisions This chapter defines the overall rules…"}, ...]\nOutput:\n[\n {"title":"Chapter 1: General Provisions","chunk_id":"1"},\n ...\n]\n\nExample 3 — Narrative + heading\nInput:\n[{"2": "This paragraph introduces the background and goals. Section 2: Definitions Key terms are explained…"}, ...]\nOutput:\n[\n {"title":"Section 2: Definitions","chunk_id":"2"},\n ...\n]\n\nExample 4 — Multiple headings in one chunk\nInput:\n[{"3": "Declarations and Commitments (I) Party B commits… (II) Party C commits… Appendix A Data Specification"}, ...]\nOutput:\n[\n {"title":"Declarations and Commitments","chunk_id":"3"},\n {"title":"(I) Party B commits","chunk_id":"3"},\n {"title":"(II) Party C commits","chunk_id":"3"},\n {"title":"Appendix A Data Specification","chunk_id":"3"},\n ...\n]\n\nExample 5 — Numbering styles\nInput:\n[{"4": "1. Scope: Defines boundaries. 2) Definitions: Terms used. III) Methods Overview."}, ...]\nOutput:\n[\n {"title":"1. Scope","chunk_id":"4"},\n {"title":"2) Definitions","chunk_id":"4"},\n {"title":"III) Methods Overview","chunk_id":"4"},\n ...\n]\n\nExample 6 — Long list (NOT headings)\nInput:\n{"5": "Item list: apples, bananas, strawberries, blueberries, mangos, peaches"}, ...]\nOutput:\n[\n {"title":"-1","chunk_id":"5"},\n ...\n]\n\nExample 7 — Mixed Chinese/English\nInput:\n{"6": "（出版信息略）This standard follows industry practices. Chapter 1: Overview 摘要… 第2节：术语与缩略语"}, ...]\nOutput:\n[\n {"title":"Chapter 1: Overview","chunk_id":"6"},\n {"title":"第2节：术语与缩略语","chunk_id":"6"},\n ...\n]"
},
{
"role": "user",
"content": "OUTPUT FORMAT\n- Return ONLY the JSON array.\n- Use double quotes.\n- No extra commentary.\n- Keep language of "title" the same as the input.\n\nINPUT\n{"1": "-11收入12650.9013.001435.97销售部178720.85\nCW-20260422026-02-12支出4190.809.00341.17采购部174529.05\nCW-20260432026-02-13收入9580.706.00522.56市场部184109.75\nCW-20260442026-02-14支出2670.606.00146.08行政部181439.15\n收入15120.8013.001700.56销售部196559.95\nCW- 20260452026-02- 15\nFAILED TO PARSE TABLE\nFAILED TO PARSE TABLE\nFAILED TO PARSE TABLE\n补充说明：1.表格共45⾏数据，刚好填满两⻚A4 PDF；2.所有数值均为财务常⽤合理数值，间隔清晰，可直接编辑修改；3.适配所有主流PDF编辑软件，下载后即可编辑。\n（注：⽂档部分内容可能由 AI ⽣成）"}"
}
]
09:49:42 - LiteLLM:INFO: utils.py:3995 -
LiteLLM completion() model= qwen3-235b-a22b; provider = dashscope
2026-06-17 09:49:42,722 INFO 33
LiteLLM completion() model= qwen3-235b-a22b; provider = dashscope
2026-06-17 09:49:42,898 INFO 33 set_progress(c53b3c9a69ee11f1b8b85533fa24cb09), progress: 0.8500000000000001, progress_msg:
2026-06-17 09:49:42,960 INFO 33 Indexing doc(测试的可编辑表格6.pdf), page(0-4), chunks(2), elapsed: 0.85
2026-06-17 09:49:45,078 INFO 33 set_progress(c53b3c9a69ee11f1b8b85533fa24cb09), progress: None, progress_msg: 09:49:43 Page(1~5): Indexing done (0.91s).
2026-06-17 09:49:45,843 INFO 33

Filtered TOC sections:
[{'title': '财务收⽀明细总表 （可编辑）', 'chunk_id': '0'}]
2026-06-17 09:49:45,845 INFO 33 [HISTORY][
{
"role": "system",
"content": "You are given a JSON array of TOC(table of contents) items. Each item has at least {"title": string} and may include an existing title hierarchical level.\n\nTask\n- For each item, assign a depth label using Arabic numerals only: top-level = 1, second-level = 2, third-level = 3, etc.\n- Multiple items may share the same depth (e.g., many 1s, many 2s).\n- Do not use dotted numbering (no 1.1/1.2). Use a single digit string per item indicating its depth only.\n- Preserve the original item order exactly. Do not insert, delete, or reorder.\n- Decide levels yourself to keep a coherent hierarchy. Keep peers at the same depth.\n\nOutput\n- Return a valid JSON array only (no extra text).\n- Each element must be {"level": "1|2|3", "title": }.\n- title must be the original title string.\n\nExamples\n\nExample A (chapters with sections)\nInput:\n["Chapter 1 Methods", "Section 1 Definition", "Section 2 Process", "Chapter 2 Experiment"]\n\nOutput:\n[\n {"level":"1","title":"Chapter 1 Methods"},\n {"level":"2","title":"Section 1 Definition"},\n {"level":"2","title":"Section 2 Process"},\n {"level":"1","title":"Chapter 2 Experiment"}\n]\n\nExample B (parts with chapters)\nInput:\n["Part I Theory", "Chapter 1 Basics", "Chapter 2 Methods", "Part II Applications", "Chapter 3 Case Studies"]\n\nOutput:\n[\n {"level":"1","title":"Part I Theory"},\n {"level":"2","title":"Chapter 1 Basics"},\n {"level":"2","title":"Chapter 2 Methods"},\n {"level":"1","title":"Part II Applications"},\n {"level":"2","title":"Chapter 3 Case Studies"}\n]\n\nExample C (plain headings)\nInput:\n["Introduction", "Background and Motivation", "Related Work", "Methodology", "Evaluation"]\n\nOutput:\n[\n {"level":"1","title":"Introduction"},\n {"level":"2","title":"Background and Motivation"},\n {"level":"2","title":"Related Work"},\n {"level":"1","title":"Methodology"},\n {"level":"1","title":"Evaluation"}\n]"
},
{
"role": "user",
"content": "['财务收⽀明细总表 （可编辑）']"
}
]
09:49:45 - LiteLLM:INFO: utils.py:3995 -
LiteLLM completion() model= qwen3-235b-a22b; provider = dashscope
2026-06-17 09:49:45,847 INFO 33
LiteLLM completion() model= qwen3-235b-a22b; provider = dashscope
2026-06-17 09:49:46,354 INFO 28 HEAD http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0 [status:200 duration:0.002s]
2026-06-17 09:49:46,360 INFO 28 POST http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0/_search [status:200 duration:0.003s]
[2026-06-17 09:49:46 +0800] [28] [INFO] 127.0.0.1:47896 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6/documents 1.1 200 27542 21360
2026-06-17 09:49:46,378 INFO 28 HEAD http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0 [status:200 duration:0.001s]
2026-06-17 09:49:46,383 INFO 28 POST http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0/_search [status:200 duration:0.003s]
[2026-06-17 09:49:46 +0800] [28] [INFO] 127.0.0.1:47912 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6/documents 1.1 200 22553 18257
[2026-06-17 09:49:46 +0800] [28] [INFO] 127.0.0.1:47916 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6 1.1 200 2360 11233
2026-06-17 09:49:46,654 INFO 33 ------------ T O C -------------
[
{
"level": "1",
"title": "财务收⽀明细总表 （可编辑）",
"chunk_id": "0"
}
]
2026-06-17 09:49:47,661 INFO 33 PUT http://192.168.80.139:1200/ragflow_105f6de65ce711f182c1edd34d87cce0/_bulk?refresh=wait_for&timeout=60s [status:200 duration:1.005s]
2026-06-17 09:49:47,804 INFO 33 set_progress(c53b3c9a69ee11f1b8b85533fa24cb09), progress: 0.9, progress_msg:
2026-06-17 09:49:48,204 INFO 33 set_progress(c53b3c9a69ee11f1b8b85533fa24cb09), progress: 1.0, progress_msg: 09:49:47 Page(1~5): Task done (19.90s)
2026-06-17 09:49:48,204 INFO 33 Chunk doc(测试的可编辑表格6.pdf), page(0-4), chunks(2), token(2654), elapsed:19.90
2026-06-17 09:49:48,206 INFO 33 handle_task done for task {"id": "c53b3c9a69ee11f1b8b85533fa24cb09", "doc_id": "c1941d8c69ee11f1b8b85533fa24cb09", "from_page": 0, "to_page": 4, "retry_count": 0, "kb_id": "7987ef80660d11f1a066912af183a8a6", "parser_id": "naive", "parser_config": {"table_context_size": 53, "image_context_size": 53, "layout_recognize": "MinerU@MinerU", "chunk_token_num": 512, "delimiter": "\n", "auto_keywords": 3, "auto_questions": 0, "html4excel": false, "topn_tags": 3, "raptor": {"use_raptor": true, "prompt": "Please summarize the following paragraphs. Be careful with the numbers, do not make things up. Paragraphs as following:\n {cluster_content}\nThe above is the content you need to summarize.", "max_token": 256, "threshold": 0.1, "max_cluster": 64, "random_seed": 0, "scope": "file", "ext": {"clustering_method": "gmm", "tree_builder": "raptor"}}, "graphrag": {"use_graphrag": true, "entity_types": ["organization", "person", "geo", "event", "category"], "method": "general", "batch_chunk_token_size": 4096, "retry_attempts": 2, "retry_backoff_seconds": 2.0, "retry_backoff_max_seconds": 60.0, "build_subgraph_timeout_per_chunk_seconds": 300, "build_subgraph_min_timeout_seconds": 600, "merge_timeout_seconds": 180, "resolution_timeout_seconds": 1800, "community_timeout_seconds": 1800, "lock_acquire_timeout_seconds": 600, "resolution": false}, "parent_child": {"use_parent_child": false, "children_delimiter": "\n"}, "children_delimiter": "", "llm_id": "qwen3-235b-a22b@Tongyi-Qianwen", "enable_children": false, "toc_extraction": true, "image_table_context_window": 53, "overlapped_percent": 0.1, "mineru_parse_method": "auto", "mineru_formula_enable": false, "mineru_table_enable": true, "mineru_lang": "Chinese", "metadata": {"type": "object", "properties": {}, "additionalProperties": false}, "built_in_metadata": [], "enable_metadata": false, "tenant_llm_id": 149}, "name": "\u6d4b\u8bd5\u7684\u53ef\u7f16\u8f91\u8868\u683c6.pdf", "type": "pdf", "location": "\u6d4b\u8bd5\u7684\u53ef\u7f16\u8f91\u8868\u683c6.pdf", "size": 202145, "tenant_id": "105f6de65ce711f182c1edd34d87cce0", "language": "Chinese", "embd_id": "text-embedding-v4@Tongyi-Qianwen", "pagerank": 0, "kb_parser_config": {"table_context_size": 53, "image_context_size": 53, "layout_recognize": "MinerU@MinerU", "chunk_token_num": 512, "delimiter": "\n", "auto_keywords": 3, "auto_questions": 0, "html4excel": false, "topn_tags": 3, "raptor": {"use_raptor": true, "prompt": "Please summarize the following paragraphs. Be careful with the numbers, do not make things up. Paragraphs as following:\n {cluster_content}\nThe above is the content you need to summarize.", "max_token": 256, "threshold": 0.1, "max_cluster": 64, "random_seed": 0, "scope": "file", "ext": {"clustering_method": "gmm", "tree_builder": "raptor"}}, "graphrag": {"use_graphrag": true, "entity_types": ["organization", "person", "geo", "event", "category"], "method": "general", "batch_chunk_token_size": 4096, "retry_attempts": 2, "retry_backoff_seconds": 2.0, "retry_backoff_max_seconds": 60.0, "build_subgraph_timeout_per_chunk_seconds": 300, "build_subgraph_min_timeout_seconds": 600, "merge_timeout_seconds": 180, "resolution_timeout_seconds": 1800, "community_timeout_seconds": 1800, "lock_acquire_timeout_seconds": 600, "resolution": false}, "parent_child": {"use_parent_child": false, "children_delimiter": "\n"}, "children_delimiter": "", "llm_id": "qwen3-235b-a22b@Tongyi-Qianwen", "enable_children": false, "toc_extraction": true, "image_table_context_window": 53, "overlapped_percent": 0.1, "mineru_parse_method": "auto", "mineru_formula_enable": false, "mineru_table_enable": true, "mineru_lang": "Chinese", "metadata": {"type": "object", "properties": {}, "additionalProperties": false}, "built_in_metadata": [], "enable_metadata": false, "tenant_llm_id": 149}, "img2txt_id": "qwen3.5-plus@Tongyi-Qianwen", "asr_id": "qwen3-asr-flash@Tongyi-Qianwen", "llm_id": "qwen3-235b-a22b@Tongyi-Qianwen", "update_time": 1781660965892, "task_type": ""}
2026-06-17 09:49:51,397 INFO 28 HEAD http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0 [status:200 duration:0.003s]
2026-06-17 09:49:51,405 INFO 28 POST http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0/_search [status:200 duration:0.004s]
[2026-06-17 09:49:51 +0800] [28] [INFO] 127.0.0.1:51920 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6/documents 1.1 200 27768 24894
2026-06-17 09:49:51,424 INFO 28 HEAD http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0 [status:200 duration:0.001s]
2026-06-17 09:49:51,429 INFO 28 POST http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0/_search [status:200 duration:0.003s]
[2026-06-17 09:49:51 +0800] [28] [INFO] 127.0.0.1:51932 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6/documents 1.1 200 22547 18080
[2026-06-17 09:49:51 +0800] [28] [INFO] 127.0.0.1:51946 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6 1.1 200 2360 11365
2026-06-17 09:49:53,136 INFO 28 HEAD http://192.168.80.139:1200/ragflow_105f6de65ce711f182c1edd34d87cce0 [status:200 duration:0.002s]
2026-06-17 09:49:53,144 INFO 28 POST http://192.168.80.139:1200/ragflow_105f6de65ce711f182c1edd34d87cce0/_search [status:200 duration:0.005s]
[2026-06-17 09:49:53 +0800] [28] [INFO] 127.0.0.1:51960 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6/documents/c1941d8c69ee11f1b8b85533fa24cb09/chunks 1.1 200 10161 21493
[2026-06-17 09:49:53 +0800] [28] [INFO] 127.0.0.1:51966 GET /api/v1/documents/c1941d8c69ee11f1b8b85533fa24cb09/preview 1.1 200 202145 11730
[2026-06-17 09:49:53 +0800] [28] [INFO] 127.0.0.1:51982 GET /api/v1/documents/images/7987ef80660d11f1a066912af183a8a6-089da1545a7357fa 1.1 200 67116 3458
[2026-06-17 09:49:53 +0800] [28] [INFO] 127.0.0.1:51984 GET /api/v1/documents/images/7987ef80660d11f1a066912af183a8a6-7e7918716d33bf39 1.1 200 16313 5142
[2026-06-17 09:49:53 +0800] [28] [INFO] 127.0.0.1:51994 GET /api/v1/documents/c1941d8c69ee11f1b8b85533fa24cb09/preview 1.1 200 202145 11940
2026-06-17 09:49:56,545 INFO 33 task_executor_28e888509d40_0 reported heartbeat: {"ip_address": "172.18.0.2", "pid": 33, "name": "task_executor_28e888509d40_0", "now": "2026-06-17T09:49:56.544+08:00", "boot_at": "2026-06-17T09:41:56.010+08:00", "pending": 0, "lag": 0, "done": 1, "failed": 0, "current": {}}
2026-06-17 09:50:26,578 INFO 33 task_executor_28e888509d40_0 reported heartbeat: {"ip_address": "172.18.0.2", "pid": 33, "name": "task_executor_28e888509d40_0", "now": "2026-06-17T09:50:26.576+08:00", "boot_at": "2026-06-17T09:41:56.010+08:00", "pending": 0, "lag": 0, "done": 1, "failed": 0, "current": {}}
2026-06-17 09:50:56,586 INFO 33 task_executor_28e888509d40_0 reported heartbeat: {"ip_address": "172.18.0.2", "pid": 33, "name": "task_executor_28e888509d40_0", "now": "2026-06-17T09:50:56.584+08:00", "boot_at": "2026-06-17T09:41:56.010+08:00", "pending": 0, "lag": 0, "done": 1, "failed": 0, "current": {}}
2026-06-17 09:51:26,593 INFO 33 task_executor_28e888509d40_0 reported heartbeat: {"ip_address": "172.18.0.2", "pid": 33, "name": "task_executor_28e888509d40_0", "now": "2026-06-17T09:51:26.591+08:00", "boot_at": "2026-06-17T09:41:56.010+08:00", "pending": 0, "lag": 0, "done": 1, "failed": 0, "current": {}}
2026-06-17 09:51:56,598 INFO 33 task_executor_28e888509d40_0 reported heartbeat: {"ip_address": "172.18.0.2", "pid": 33, "name": "task_executor_28e888509d40_0", "now": "2026-06-17T09:51:56.596+08:00", "boot_at": "2026-06-17T09:41:56.010+08:00", "pending": 0, "lag": 0, "done": 1, "failed": 0, "current": {}}
2026-06-17 09:52:26,606 INFO 33 task_executor_28e888509d40_0 reported heartbeat: {"ip_address": "172.18.0.2", "pid": 33, "name": "task_executor_28e888509d40_0", "now": "2026-06-17T09:52:26.605+08:00", "boot_at": "2026-06-17T09:41:56.010+08:00", "pending": 0, "lag": 0, "done": 1, "failed": 0, "current": {}}
2026-06-17 09:52:56,612 INFO 33 task_executor_28e888509d40_0 reported heartbeat: {"ip_address": "172.18.0.2", "pid": 33, "name": "task_executor_28e888509d40_0", "now": "2026-06-17T09:52:56.611+08:00", "boot_at": "2026-06-17T09:41:56.010+08:00", "pending": 0, "lag": 0, "done": 1, "failed": 0, "current": {}}
2026-06-17 09:53:26,619 INFO 33 task_executor_28e888509d40_0 reported heartbeat: {"ip_address": "172.18.0.2", "pid": 33, "name": "task_executor_28e888509d40_0", "now": "2026-06-17T09:53:26.617+08:00", "boot_at": "2026-06-17T09:41:56.010+08:00", "pending": 0, "lag": 0, "done": 1, "failed": 0, "current": {}}
2026-06-17 09:53:46,579 INFO 28 HEAD http://192.168.80.139:1200/ragflow_105f6de65ce711f182c1edd34d87cce0 [status:200 duration:0.002s]
2026-06-17 09:53:46,589 INFO 28 POST http://192.168.80.139:1200/ragflow_105f6de65ce711f182c1edd34d87cce0/_search [status:200 duration:0.007s]
2026-06-17 09:53:46,602 INFO 28 HEAD http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0 [status:200 duration:0.001s]
2026-06-17 09:53:46,610 INFO 28 POST http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0/_search [status:200 duration:0.006s]
2026-06-17 09:53:46,624 INFO 28 HEAD http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0 [status:200 duration:0.001s]
2026-06-17 09:53:46,631 INFO 28 POST http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0/_search [status:200 duration:0.005s]
[2026-06-17 09:53:46 +0800] [28] [INFO] 127.0.0.1:37486 GET /api/v1/users/me 1.1 200 521 89727
[2026-06-17 09:53:46 +0800] [28] [INFO] 127.0.0.1:37498 GET /api/v1/users/me/models 1.1 200 592 90156
[2026-06-17 09:53:46 +0800] [28] [INFO] 127.0.0.1:37516 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6/graph 1.1 200 45 90480
[2026-06-17 09:53:46 +0800] [28] [INFO] 127.0.0.1:37506 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6/documents 1.1 200 27768 90802
[2026-06-17 09:53:46 +0800] [28] [INFO] 127.0.0.1:37520 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6/documents 1.1 200 22547 91128
[2026-06-17 09:53:46 +0800] [28] [INFO] 127.0.0.1:37530 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6 1.1 200 2360 91651
2026-06-17 09:53:46,657 WARNING 28 Database connection issue (attempt 1/5): (0, '')
2026-06-17 09:53:46,673 INFO 28 HEAD http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0 [status:200 duration:0.002s]
2026-06-17 09:53:46,678 INFO 28 POST http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0/_search [status:200 duration:0.003s]
[2026-06-17 09:53:46 +0800] [28] [INFO] 127.0.0.1:37546 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6/documents 1.1 200 22547 22205
[2026-06-17 09:53:47 +0800] [28] [INFO] 127.0.0.1:37544 GET /api/v1/tenants 1.1 200 241 1010452
[2026-06-17 09:53:51 +0800] [28] [INFO] 127.0.0.1:48994 GET /api/v1/datasets 1.1 200 26874 26067
[2026-06-17 09:53:51 +0800] [28] [INFO] 127.0.0.1:48998 GET /api/v1/datasets 1.1 200 26874 12789
[2026-06-17 09:53:51 +0800] [28] [INFO] 127.0.0.1:49000 GET /api/v1/users/me 1.1 200 521 5507
[2026-06-17 09:53:52 +0800] [28] [INFO] 127.0.0.1:49012 GET /api/v1/users/me 1.1 200 521 77409
[2026-06-17 09:53:52 +0800] [28] [INFO] 127.0.0.1:49020 GET /api/v1/users/me/models 1.1 200 592 77784
[2026-06-17 09:53:52 +0800] [28] [INFO] 127.0.0.1:49034 GET /api/v1/datasets 1.1 200 26874 78301
[2026-06-17 09:53:52 +0800] [28] [INFO] 127.0.0.1:49042 GET /v1/llm/list 1.1 200 19373 78466
[2026-06-17 09:53:52 +0800] [28] [INFO] 127.0.0.1:49006 GET /api/v1/chats 1.1 200 13787 92350
[2026-06-17 09:53:52 +0800] [28] [INFO] 127.0.0.1:49022 GET /api/v1/tenants 1.1 200 242 79870
[2026-06-17 09:53:52 +0800] [28] [INFO] 127.0.0.1:49050 GET /api/v1/users/me 1.1 200 521 5444
[2026-06-17 09:53:53 +0800] [28] [INFO] 127.0.0.1:49052 GET /api/v1/users/me 1.1 200 521 13080
[2026-06-17 09:53:53 +0800] [28] [INFO] 127.0.0.1:49066 GET /api/v1/users/me/models 1.1 200 592 15937
[2026-06-17 09:53:53 +0800] [28] [INFO] 127.0.0.1:49076 GET /api/v1/tenants 1.1 200 242 8379
[2026-06-17 09:53:53 +0800] [28] [INFO] 127.0.0.1:49090 GET /api/v1/system/version 1.1 200 48 1286
[2026-06-17 09:53:53 +0800] [28] [INFO] 127.0.0.1:49092 GET /api/v1/connectors 1.1 200 41 7631
2026-06-17 09:53:56,625 INFO 33 task_executor_28e888509d40_0 reported heartbeat: {"ip_address": "172.18.0.2", "pid": 33, "name": "task_executor_28e888509d40_0", "now": "2026-06-17T09:53:56.623+08:00", "boot_at": "2026-06-17T09:41:56.010+08:00", "pending": 0, "lag": 0, "done": 1, "failed": 0, "current": {}}
2026-06-17 09:54:26,744 INFO 33 task_executor_28e888509d40_0 reported heartbeat: {"ip_address": "172.18.0.2", "pid": 33, "name": "task_executor_28e888509d40_0", "now": "2026-06-17T09:54:26.742+08:00", "boot_at": "2026-06-17T09:41:56.010+08:00", "pending": 0, "lag": 0, "done": 1, "failed": 0, "current": {}}
2026-06-17 09:54:56,751 INFO 33 task_executor_28e888509d40_0 reported heartbeat: {"ip_address": "172.18.0.2", "pid": 33, "name": "task_executor_28e888509d40_0", "now": "2026-06-17T09:54:56.749+08:00", "boot_at": "2026-06-17T09:41:56.010+08:00", "pending": 0, "lag": 0, "done": 1, "failed": 0, "current": {}}
[2026-06-17 09:55:01 +0800] [28] [INFO] 127.0.0.1:45090 GET /api/v1/datasets 1.1 200 26874 25549
[2026-06-17 09:55:01 +0800] [28] [INFO] 127.0.0.1:45096 GET /api/v1/datasets 1.1 200 26874 25947
[2026-06-17 09:55:02 +0800] [28] [INFO] 127.0.0.1:45102 GET /api/v1/users/me 1.1 200 521 5576
2026-06-17 09:55:03,674 INFO 28 HEAD http://192.168.80.139:1200/ragflow_105f6de65ce711f182c1edd34d87cce0 [status:200 duration:0.002s]
2026-06-17 09:55:03,679 INFO 28 POST http://192.168.80.139:1200/ragflow_105f6de65ce711f182c1edd34d87cce0/_search [status:200 duration:0.002s]
2026-06-17 09:55:03,893 INFO 28 HEAD http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0 [status:200 duration:0.001s]
2026-06-17 09:55:03,898 INFO 28 POST http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0/_search [status:200 duration:0.002s]
[2026-06-17 09:55:03 +0800] [28] [INFO] 127.0.0.1:45106 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6/graph 1.1 200 45 235751
[2026-06-17 09:55:03 +0800] [28] [INFO] 127.0.0.1:45120 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6/documents 1.1 200 27768 236237
2026-06-17 09:55:03,921 INFO 28 HEAD http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0 [status:200 duration:0.001s]
2026-06-17 09:55:03,926 INFO 28 POST http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0/_search [status:200 duration:0.002s]
[2026-06-17 09:55:03 +0800] [28] [INFO] 127.0.0.1:45130 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6 1.1 200 2360 26975
[2026-06-17 09:55:03 +0800] [28] [INFO] 127.0.0.1:45146 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6/documents 1.1 200 22547 27291
2026-06-17 09:55:03,941 INFO 28 HEAD http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0 [status:200 duration:0.001s]
2026-06-17 09:55:03,945 INFO 28 POST http://192.168.80.139:1200/ragflow_doc_meta_105f6de65ce711f182c1edd34d87cce0/_search [status:200 duration:0.002s]
[2026-06-17 09:55:03 +0800] [28] [INFO] 127.0.0.1:45154 GET /api/v1/datasets/7987ef80660d11f1a066912af183a8a6/documents 1.1 200 22547 16495

[dosubot]

Hi @yuwei88888! I'm Dosu and I'm helping the InfiniFlow team.

The "FAILED TO PARSE TABLE" message is a fallback placeholder generated by RAGFlow when MinerU's table extraction returns empty content. It originates in deepdoc/parser/mineru_parser.py in the _transfer_to_sections() method [1]:

case MinerUContentType.TABLE:
    section = output.get("table_body", "") + "\n".join(output.get("table_caption", [])) + "\n".join(
        output.get("table_footnote", []))
    if not section.strip():
        section = "FAILED TO PARSE TABLE"

What's happening: MinerU's layout detection correctly identifies table regions in your PDF (you can see it parsed 8 blocks), but the table content extraction model returns empty table_body/table_caption/table_footnote fields for some of those table blocks. This is a MinerU-side extraction issue — particularly common with dense financial tables spanning multiple pages [2].

Workarounds to try:

1. Switch mineru_parse_method to "ocr" — may improve extraction for tables that the default "auto" method struggles with.
2. Use a different layout recognition engine — try setting layout_recognize to "Plain Text" or use the "Naive" parser, which bypasses layout analysis entirely and may handle simple tabular structures better.
3. Pre-process the PDF — splitting the 2-page table into single-page PDFs or converting to a non-editable format before uploading may help MinerU extract the content.
4. Check your MinerU server version — there have been compatibility improvements in recent RAGFlow releases (e.g., MinerU 3.x support in PR Fix: MinerU 3.x output discovery and API contract #14282). Ensure your MinerU API server at http://101.36.73.92:8091 is running a compatible version [3].

This appears related to a broader issue reported in GitHub issue #16060 where MinerU returns empty arrays for table content in content_list.json [2].

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.

[yuwei88888]

@dosu
I understand the problem caused by the empty array that you mentioned. But why couldn't the normal table be parsed successfully? The analysis has been converted into text.

[dosubot]

Good question. There are actually two separate issues happening with your table:

Issue 1 — "FAILED TO PARSE TABLE": For some table blocks, MinerU returned empty table_body content, so RAGFlow inserted the placeholder. This means MinerU's table extraction model couldn't extract those specific regions at all.

Issue 2 — Why the successfully extracted rows appear as text, not a table: This is by design in the MinerU parser pipeline. Even when MinerU successfully extracts table HTML (the rows you can see in the chunks), RAGFlow's _sanitize_section_text() method deliberately strips all HTML structure and converts it to plain text [1]:

• <br> tags → newlines
• </tr>, </td>, </table> → newlines
• All remaining HTML tags → removed entirely

So the data flow is: MinerU extracts HTML table → RAGFlow sanitizes to plain text → chunks stored as text. The _transfer_to_tables() method is currently unimplemented (returns an empty list), meaning tables are never preserved as structured objects in the MinerU parser path [2].

In short: Your table data was extracted by MinerU for most rows, but the current MinerU parser architecture converts all table HTML to plain text for indexing purposes. There is no code path that preserves formatted table output when using MinerU as the layout engine.

If you need structured table output, you could try:

• Using RAGFlow's DeepDoc layout recognition instead of MinerU — it has a dedicated TableStructureRecognizer that generates proper HTML tables [3]
• Using the Markdown parser if your content can be converted to markdown format first

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.

[yuwei88888]

@dosu
What you mean is that only when mineru parses the tables within the PDF file, does it return the table content in the form of an MD format table? Then can ragflow parse it normally?

[dosubot]

Not quite. Let me clarify:

MinerU returns table content as HTML (with <tr>, <td>, <br> tags), not Markdown — even though your request has return_md: True [1]. The return_md parameter affects other content types but does not change the table format.

RAGFlow then strips all HTML and converts it to plain text — it does not render the table visually. So the flow is:

1. MinerU detects a table region → extracts it as HTML in table_body
2. RAGFlow's _sanitize_section_text() converts </tr>, </td> → newlines, then removes all remaining HTML tags [2]
3. The result is plain text stored in chunks

In your case, the problem is that for some table blocks, MinerU returned empty table_body (hence "FAILED TO PARSE TABLE"), while for other blocks it returned HTML which RAGFlow converted to the plain text you see in your chunks. Either way, you won't get a formatted table in the output — that's the current architectural limitation of the MinerU parser path.

To summarize: The issue isn't about MD vs HTML format. The issue is:

• MinerU failed to extract content for some table regions entirely (empty table_body)
• Even for successfully extracted regions, RAGFlow flattens the HTML to plain text by design

If you need structured/visual table output, you'd need to use the DeepDoc layout recognizer instead of MinerU — it has a dedicated table structure recognizer that generates proper HTML tables for display [3].

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.

[yuwei88888]

@dosu
Will you optimize this part later? Make it compatible with the HTML format returned by Mineru, and cut it into a table format normally. We can't just remove all the HTML tags returned. This approach is too crude.

[dosubot]

Your feedback is valid — stripping all HTML tags from MinerU's table output does lose valuable structure.

There is actually an open PR (#16118) that directly addresses this concern [1]. It proposes conditional sanitization — preserving HTML table structure when table recognition is enabled, rather than unconditionally stripping all tags. This aligns with exactly what you're asking for.

Additionally, maintainer comments indicate that MinerU's HTML table format is considered the preferred format for retrieval, and the intention is to keep that structure intact rather than flatten it [2]. The aggressive stripping in PR #14920 was focused on cleaning up stray HTML fragments, but it inadvertently affected valid table structure too [3].

That said, the _transfer_to_tables() method (which would properly handle tables as separate structured objects) remains unimplemented since October 2025 [4], so full table-as-structured-data support is not yet on the immediate roadmap.

In the short term, you can track PR #16118 for the conditional preservation fix. For now, your best option for proper table output remains using the DeepDoc layout recognizer instead of MinerU.

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.

[yuwei88888]

@dosu
Yesterday I tried that branch, but there was actually a problem with it. It didn't treat the table as a single chunk but instead divided it into sections based on a fixed number of characters. This resulted in the table being split into two parts, and the latter part would directly display the label format because it lacked the

tag.If I want to correct this, could you offer me any suggestions?

[yuwei88888]

@dosu
Would there be any other issues if I reused the processing approach of DeepDoc?

[dosubot]

Reusing DeepDoc's approach is feasible but there are a few things to watch out for:

What DeepDoc does differently: It marks tables with ck_type: "table", which causes the _merge_cks() function in naive.py to treat them as indivisible chunks — they completely bypass token-based merging/splitting [1]. This is exactly why DeepDoc tables never get cut in half.

Potential issues if you adapt this for MinerU:

1. Position coordinate format mismatch — MinerU embeds coordinates as text tags (@@page\tx0\tx1\ttop\tbottom##) within section text [2], while DeepDoc passes them directly as objects. You'd need to extract these coordinates before creating the table chunk, or preserve the tags inside the HTML content.

2. HTML quality differences — DeepDoc generates its own HTML via TableStructureRecognizer [3] with guaranteed well-formed structure. MinerU's table_body comes from an external API and may have inconsistencies. You should validate the HTML is well-formed before storing it as a chunk.

3. Image thumbnails — DeepDoc uses a tokenize_table() pipeline with direct bbox access, while MinerU relies on deferred crop() with embedded position tags [4]. You may need to handle table_img_path from MinerU's output or adapt the cropping logic.

4. Large tables exceeding token limits — DeepDoc stores tables whole regardless of token count. If MinerU extracts very large tables (like your 45-row financial table), the chunk could be much larger than chunk_token_num. This won't break anything functionally, but may affect retrieval relevance for embedding models with token limits.

Suggested implementation path:

In _transfer_to_sections(), when you encounter a MinerUContentType.TABLE block, instead of concatenating and sanitizing, return the raw HTML with a marker that tells the chunking pipeline to treat it as a table chunk (similar to how DeepDoc returns (text, image, table) tuples). Then ensure it gets ck_type: "table" so _merge_cks() won't split it.

The cleanest approach would be to implement _transfer_to_tables() [5] to actually return table objects, then feed those into the same chunking path DeepDoc uses.

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.

[yuwei88888]

@dosu Thank you for your help. I'll give it another try.

[dosubot]

You're welcome! Good luck with the implementation. Feel free to follow up here if you run into any issues or need further guidance.

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.