Output of ONNX inference model for ch_PP-OCRv3_rec format conversion, how?

[nestorprojects]

Hi,
I converted using Paddle2ONNX inference model mentioned in the title, which when tested using PaddleOCR.ocr function for example image I used returns 40 lines of text with predicted bounding box and Texts with probabilities within them,
When using ONNX model, it returns 40 arrays of size 6625 (which probably somehow correlate to 6623 characters), now first one and last two are with some numbers (like image in attachment), how do I get what ocr function produced, and displayed from that array, if actually that's how it can work?