fix onnx in latexocr

[liuhongen1234567]

No description provided.

[paddle-bot]

Thanks for your contribution!

[Copilot]

Pull Request Overview

This PR addresses issues with ONNX export in the LaTeX OCR head by simplifying the sampling logic during inference.

• Modified sampling by replacing multinomial sampling with argmax in the generate and generate_export functions.
• Removed the branch that applied filter_logits_fn for top_k and top_p filtering.

Comments suppressed due to low confidence (2)

ppocr/modeling/heads/rec_latexocr_head.py:910

• Replacing softmax and multinomial sampling with argmax disregards the temperature and filtering parameters, which may alter the intended prediction distribution. Consider documenting this change or refactoring the unused parameters if deterministic output is acceptable for ONNX export.

sample = paddle.argmax(logits, axis=1).reshape([-1, 1])

ppocr/modeling/heads/rec_latexocr_head.py:962

• The switch to a deterministic argmax approach in generate_export may impact the output behavior compared to the previous stochastic sampling method. Ensure this change aligns with the overall model requirements and ONNX export compatibility.

sample = paddle.argmax(logits, axis=1).reshape([-1, 1])