pythera/sati

Model

The segment-any-text/sat-12l-sm model is a part of the Segment Any Text (SAT) framework, specifically utilized within the WTPSplit project. This model enables efficient and flexible text segmentation by leveraging XML-Roberta architectures.

Available Model Types

We provide multiple formats of the sat-12l-sm model:

ONNX

1. Default: Standard ONNX model for general use.
2. Optimized: Performance-enhanced version with improved inference speed.

TensorRT

1. FP32: Floating-point precision model, ensuring high accuracy.
2. FP32+INT8: Mixed precision (FP32 and INT8) model for optimized inference speed with reduced memory footprint.

Reference

1. Github
2. Library and the WtP models