Dataset Preview

Split (2)

train

The full dataset viewer is not available (click to read why). Only showing a preview of the rows.

Previous task error: I/O error: I/O error: Permission denied (os error 13)

Error code:   UnexpectedError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

images list	conversations list	source_file string
["/9j/4AAQSkZJRgABAQEAlgCWAAD/2wBDAAIBAQEBAQIBAQECAgICAgQDAgICAgUEBAMEBgUGBgYFBgYGBwkIBgcJBwYGCAsICQ(...TRUNCATED)	[{"role":"user","content":"<image>\nTASK: Extract all trucking load fields from this rate confirmati(...TRUNCATED)	ELI52018LoadTender-EXP7389228.pdf
["/9j/4AAQSkZJRgABAQEAlgCWAAD/2wBDAAIBAQEBAQIBAQECAgICAgQDAgICAgUEBAMEBgUGBgYFBgYGBwkIBgcJBwYGCAsICQ(...TRUNCATED)	[{"role":"user","content":"<image>\nTASK: Extract all trucking load fields from this rate confirmati(...TRUNCATED)	BillOfLading (4).pdf
["/9j/4AAQSkZJRgABAQEAlgCWAAD/2wBDAAIBAQEBAQIBAQECAgICAgQDAgICAgUEBAMEBgUGBgYFBgYGBwkIBgcJBwYGCAsICQ(...TRUNCATED)	[{"role":"user","content":"<image>\n<image>\nTASK: Extract all trucking load fields from this rate c(...TRUNCATED)	PayQuote_E857601_2026-03-25_1137.pdf
["/9j/4AAQSkZJRgABAQEAlgCWAAD/2wBDAAIBAQEBAQIBAQECAgICAgQDAgICAgUEBAMEBgUGBgYFBgYGBwkIBgcJBwYGCAsICQ(...TRUNCATED)	[{"role":"user","content":"<image>\n<image>\nTASK: Extract all trucking load fields from this rate c(...TRUNCATED)	Purchase Order_SD-29794 (1).pdf
["/9j/4AAQSkZJRgABAQEAlgCWAAD/2wBDAAIBAQEBAQIBAQECAgICAgQDAgICAgUEBAMEBgUGBgYFBgYGBwkIBgcJBwYGCAsICQ(...TRUNCATED)	[{"role":"user","content":"<image>\n<image>\n<image>\nTASK: Extract all trucking load fields from th(...TRUNCATED)	Carrier Rate confirmation-14.pdf
["/9j/4AAQSkZJRgABAQEAlgCWAAD/2wBDAAIBAQEBAQIBAQECAgICAgQDAgICAgUEBAMEBgUGBgYFBgYGBwkIBgcJBwYGCAsICQ(...TRUNCATED)	[{"role":"user","content":"<image>\n<image>\nTASK: Extract all trucking load fields from this rate c(...TRUNCATED)	31882401_rate_agreement_doc_1763747621832.pdf
["/9j/4AAQSkZJRgABAQEAlgCWAAD/2wBDAAIBAQEBAQIBAQECAgICAgQDAgICAgUEBAMEBgUGBgYFBgYGBwkIBgcJBwYGCAsICQ(...TRUNCATED)	[{"role":"user","content":"<image>\n<image>\n<image>\nTASK: Extract all trucking load fields from th(...TRUNCATED)	05.13 ID-ID.pdf
["/9j/4AAQSkZJRgABAQEAlgCWAAD/2wBDAAIBAQEBAQIBAQECAgICAgQDAgICAgUEBAMEBgUGBgYFBgYGBwkIBgcJBwYGCAsICQ(...TRUNCATED)	[{"role":"user","content":"<image>\n<image>\nTASK: Extract all trucking load fields from this rate c(...TRUNCATED)	1084136 (2).pdf
["/9j/4AAQSkZJRgABAQEAlgCWAAD/2wBDAAIBAQEBAQIBAQECAgICAgQDAgICAgUEBAMEBgUGBgYFBgYGBwkIBgcJBwYGCAsICQ(...TRUNCATED)	[{"role":"user","content":"<image>\n<image>\n<image>\n<image>\nTASK: Extract all trucking load field(...TRUNCATED)	semail (6).pdf
["/9j/4AAQSkZJRgABAQEAlgCWAAD/2wBDAAIBAQEBAQIBAQECAgICAgQDAgICAgUEBAMEBgUGBgYFBgYGBwkIBgcJBwYGCAsICQ(...TRUNCATED)	[{"role":"user","content":"<image>\n<image>\nTASK: Extract all trucking load fields from this rate c(...TRUNCATED)	Carrier_rate_confirmation_-__31426-42761.pdf

End of preview.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

# finetune_ocr

Генерация датасета для дообучения OCR-модели на парсинг.rate confirmation документов.

Зачем

Парсер в боте (NVIDIA API) работает отлично, но:

Долгий (2-5 сек на документ)
Зависимость от NVIDIA (неизвестно сколько проработает)
Дорогой при большом объёме

Цель: дообучить быструю локальную модель на размеченных данных.

Структура

finetune_ocr/
├── config.py              # Настройки (сервер, таймауты, пути)
├── rc_parser.py           # Парсер (адаптирован из бота, работает с локальным сервером)
├── fairy_vision.py        # OCR через Tesseract (фоллбэк)
├── gather_dataset.py      # Главный скрипт — обрабатывает PDF и сохраняет JSON
├── requirements.txt
├── rate_confirmations/    # Сюда складывать PDF для разметки
│   └── (46 файлов скопировано из newbot/test_rcs)
└── dataset/               # Сюда сохраняются JSON-ответы

Использование

# Установка зависимостей
pip install -r requirements.txt

# Показать доступные модели на сервере
python gather_dataset.py --list-models

# Обработать все новые файлы
python gather_dataset.py

# Обработать конкретный файл
python gather_dataset.py --file "RateConfirmation.pdf"

# Перес processing всех файлов
python gather_dataset.py --reprocess

# Указать конкретную модель
python gather_dataset.py --model "qwen/qwen3.5-122b-a10b"

Формат датасета

Каждый JSON в dataset/:

{
  "source_file": "RateConfirmation.pdf",
  "parsed": {
    "origin": "CORONA CA",
    "dest": "CLAYPOOL AZ",
    "pu_date": "2026-01-19",
    "del_date": "2026-01-20",
    "document_date": "2026-01-18",
    "total_rate": 1700.0,
    "weight": "48000",
    "commodity": "CONSTRUCTION",
    "pickup_num": "1009340",
    "instructions": "Tarp required",
    "stops": [
      {"raw_type": "PU 1", "type": "PICKUP", "address": "...", "date": "...", "time": "..."}
    ]
  }
}

Требования

Локальный llama-server (Tailscale: 100.75.222.32:8081 или localhost:8080)
Python 3.11+
PyMuPDF, httpx, Pillow, pytesseract

Downloads last month: 286

Total file size:

490 MB