Upload README.md with huggingface_hub

README.md

@@ -13,23 +13,6 @@ tags:
 - prime-environments
 size_categories:
 - n<1K
-configs:
-- config_name: default
-  data_files:
-  - split: test
-    path: data/test-*
-dataset_info:
-  features:
-  - name: images
-    list: image
-  - name: output
-    dtype: string
-  splits:
-  - name: test
-    num_bytes: 699664919
-    num_examples: 420
-  download_size: 649638911
-  dataset_size: 699664919
 ---
 
 # MPDocBench-Parse for RLVR (images → gold Markdown)
@@ -38,9 +21,8 @@ A Hugging Face repackage of **MPDocBench-Parse** (*Benchmarking Practical
 Multi-page Document Parsing*, arXiv:2605.22100; USTC + Tongyi Lab, Alibaba),
 built for RL training/evaluation with Prime Intellect Environments (`verifiers`).
 
-- **Input**: `images` — the ordered page images of one document, plus
-  `prompt_text` — the parsing instruction.
-- **Output**: `gold_markdown` — the benchmark ground truth reconstructed into
+- **Input**: `images` — the ordered page images of one document.
+- **Output**: `output` — the benchmark ground truth reconstructed into
   the single Markdown string the official evaluator expects a parser to
   produce: reading order across pages, `#` heading levels from the
   parent/son heading tree, HTML `<table>` tables, LaTeX display formulas
@@ -54,32 +36,32 @@ built for RL training/evaluation with Prime Intellect Environments (`verifiers`)
 
 ## Split
 
-A single `test` split with all 409 documents. MPDocBench-Parse is a
-benchmark and has no official train split.
+A single `test` split with all 420 documents. MPDocBench-Parse is a
+benchmark and has no official train split. Page images for the 11
+`pptSlides` documents come from the gated `NTT-hil-insight/SlideVQA`
+dataset, per upstream's `tools/download_data_from_slidevqa.py`.
 
 ## Columns
 
-`doc_id`, `data_source` (15 document types), `language` (`english`/`chinese`),
-`num_pages`, `images` (list of page images), `prompt_text`, `gold_markdown`,
-and gold element counts `n_tables` / `n_figures` / `n_formulas` / `n_headings`
-/ `gold_chars` for curriculum filtering.
+Exactly two: `images` (list of page images, in order) and `output`
+(the gold Markdown string).
 
 ## Use with Prime Environments
 
 ```python
 from datasets import load_dataset
 ds = load_dataset("albertklorer/mpdocbench")
-print(ds["test"][0]["gold_markdown"][:500])
+print(ds["test"][0]["output"][:500])
 ```
 
 The companion environment `environments/mpdocbench_parse/` maps rows to a
 multimodal `verifiers` prompt (one `image_url` part per page) with
-`answer=gold_markdown` and a composite reward approximating the benchmark's
+`answer=output` and a composite reward approximating the benchmark's
 8 tasks (text NED, table, formula, figure F1, heading hierarchy).
 
 ## Validation of the gold reconstruction
 
-Feeding `gold_markdown` back as a "prediction" to the official upstream
+Feeding `output` back as a "prediction" to the official upstream
 evaluator (15 documents, one per document type; CDM skipped — no LaTeX
 rendering stack): FigureF1 1.000, table TEDS 1.000, merged-table TEDS
 1.000, merged-text edit distance 0.000, text edit distance (whole) 0.003,