Threshold issues when using translation models

[loongshuai]

When I used the author's published models for text translation, I found some problems and needed to adjust the threshold when there were too many words. But I'm not sure what kind of threshold is suitable for what size of text. For example, in the translation of the following text:

"You've got to be clear about what is the sort of sector where we can promote, cooperate; and ones, frankly, where we can't. I wouldn't personally bring a Chinese company into our steel sector. I think steel is a very sensitive area," he said.
  Jingye bought British Steel from the government in 2020 after the company became insolvent.
  Since coming to office in 2024, the Labour government has stepped up engagement with China after tensions under previous 
  Conservative governments over human rights, Hong Kong and latterly restrictions on investment over security concerns.
  Reynolds said he viewed other sectors such as car making, life sciences and agricultural products as less sensitive areas for Chinese investment.
  British finance minister Rachel Reeves visited Beijing in January and Chinese foreign minister Wang Yi visited London in February to revive talks that were paused for over six years.

I use 50 which results in an incomplete translation, and using 80 is just fine.

[niedev]

Hi, you need to find the minimum threshold that prevents problems with longer sentences, because the less you split the phrase the better (the model will have more context for the translation), I found 200 to be the sweet spot bot if you have some problems you can decrease the threshold little by little until the final output is good.