Upload folder using huggingface_hub

README.md

@@ -1,3 +1,102 @@
----
-license: pddl
----
+---
+license: mit
+task_categories:
+  - text-generation
+language:
+  - en
+size_categories:
+  - 10K<n<100K
+tags:
+  - ocr
+  - document-processing
+  - deepseek
+configs:
+  - config_name: epstein_estate_2025_09
+    data_files:
+      - split: train
+        path: epstein_estate_2025_09-train.parquet
+  - config_name: epstein_estate_2025_11
+    data_files:
+      - split: train
+        path: epstein_estate_2025_11-train.parquet
+  - config_name: house_doj_2025_09
+    data_files:
+      - split: train
+        path: house_doj_2025_09-train.parquet
+dataset_info:
+  features:
+    - name: source_file
+      dtype: string
+    - name: page_number
+      dtype: int32
+    - name: split
+      dtype: string
+    - name: text
+      dtype: string
+    - name: ocr_model
+      dtype: string
+    - name: resolution
+      dtype: string
+    - name: char_count
+      dtype: int32
+    - name: word_count
+      dtype: int32
+    - name: line_count
+      dtype: int32
+    - name: is_empty
+      dtype: bool
+---
+
+# epstractor-ocr
+
+This dataset contains OCR-extracted text from documents processed using DeepSeek-OCR.
+
+## Dataset Summary
+
+- **Total Records**: 56,774
+- **Total Source Files**: 56,774
+- **Format**: Parquet (Snappy compression)
+
+## Dataset Structure
+
+### Data Fields
+
+- `source_file`: Original source filename (without page numbers or extensions)
+- `page_number`: Page number for multi-page documents (null for single-page images)
+- `split`: Dataset split (train/test/validation)
+- `text`: OCR-extracted text in markdown format
+- `ocr_model`: Model used for OCR extraction (e.g., "deepseek-ai/DeepSeek-OCR")
+- `resolution`: Resolution mode used during OCR (tiny/small/base/large/gundam)
+- `char_count`: Number of characters in the extracted text
+- `word_count`: Number of words in the extracted text
+- `line_count`: Number of lines in the extracted text
+- `is_empty`: Boolean indicating if the page is essentially blank (<10 non-whitespace chars)
+
+### Subsets
+
+- **epstein_estate_2025_09**: 356 records from 356 files
+- **epstein_estate_2025_11**: 23,124 records from 23,124 files
+- **house_doj_2025_09**: 33,294 records from 33,294 files
+
+## Usage
+
+```python
+from datasets import load_dataset
+
+# Load entire dataset
+dataset = load_dataset("path/to/dataset")
+
+# Load specific subset
+dataset = load_dataset("path/to/dataset", name="subset_name")
+
+# Load with streaming for large datasets
+dataset = load_dataset("path/to/dataset", streaming=True)
+```
+
+## Source
+
+This dataset was created using the OCR Pipeline with DeepSeek-OCR and vLLM.
+
+## License
+
+MIT License

epstein_estate_2025_09-train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5748c9f1761c290a64fb45e33a92eb191744ade53e364ee8a66b13ab09671446
+size 242955

epstein_estate_2025_11-train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:019aee1f9970c28f3101c9c942e2fee239b63a6f8a944312a3174714ee5e9179
+size 29235706

house_doj_2025_09-train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca65ff8d1ccac8fcf014acec14a58cfb7c46cbeb3dca00fbf4564206d517ff95
+size 28763058