Fine-tune summarization never works well

Hello, guys. I hope you are all good.

I am a beginner to Hugging Face, and I am trying to fine-tune a model for a summarization task.

I tried bert-multilingual-cased and mT5, but I have the same problem with both of them: the summary does not exceed about 11 words, and they are not good summaries.

I used the SAMSum conversational dataset translated into Arabic. I do not know how to solve this problem! I thought by specifying
max_input_length = 1024 and max_target_length = 128, I would guarantee it works, but it does not, and I cannot figure out where the problem is. You can find the code (I copied a notebook I found from a YouTube video) for fine-tuning here:

Could you please help me find the problem?

you can try my model here