Spaces:
Running
Running
| title: Russian Contract Extractor | |
| emoji: ๐ | |
| colorFrom: blue | |
| colorTo: indigo | |
| sdk: gradio | |
| sdk_version: 6.20.0 | |
| app_file: app.py | |
| python_version: "3.12" | |
| startup_duration_timeout: 1h | |
| short_description: Extract structured JSON from Russian contracts | |
| models: | |
| - zieglerd/RussianConcsExt | |
| # Russian Contract Extractor | |
| Interactive demo for | |
| [`zieglerd/RussianConcsExt`](https://huggingface.co/zieglerd/RussianConcsExt), | |
| a LoRA adapter that extracts structured JSON from Russian procurement and | |
| corporate contracts. | |
| The adapter metadata identifies `Qwen/Qwen2.5-32B-Instruct` as its base model. | |
| Inference uses 4-bit NF4 quantization on ZeroGPU. | |
| ## Limitations | |
| - Optimized for contracts with one supplier. | |
| - Extraction quality depends on OCR quality and document complexity. | |
| - Long inputs are truncated to 16,384 tokens in this demo. | |
| - Results must be reviewed before legal or financial use. | |
| - This tool does not provide legal advice. | |