| language: | |
| - en | |
| tags: | |
| - text2text-generation | |
| - onnx | |
| - int8 | |
| - legal | |
| - amicus | |
| license: mit | |
| widget: | |
| - text: "Humanize this legal text: The party of the first part shall herewith indemnify..." | |
| # Amicus Humanizer v1 (ONNX INT8) | |
| This is the highly optimized, CPU-ready ONNX INT8 quantized version of the **Amicus Humanizer v1** model. | |
| It has been specifically fine-tuned and quantized for the Dockase ecosystem to rewrite dense, complex legal jargon into plain, easily understandable English. | |
| ## Model Details | |
| - **Architecture:** Seq2Seq (Encoder-Decoder) based on FLAN-T5 | |
| - **Task:** Text-to-Text Generation (Legal text simplification) | |
| - **Format:** ONNX | |
| - **Quantization:** INT8 (Dynamic Quantization with AVX512 optimization) | |
| - **Target Deployment:** Fast inference on CPU servers | |
| ## Usage | |
| This model is designed to be run using `optimum.onnxruntime` or raw ONNX Runtime inference servers. | |