Bad Programming Performance

#4
by JosephH - opened

Hi Trendyol team,
I hope you doing well. Congratulations on your robust LLM. I hope Allah leads you better success.
I have tested your model on colab and it's programming capability wasn't good as much as I expected.
It rejects to write code even though I believe it should be able to. Maybe some prompt technique might reveal its hidden gems.
Even though I haven't tested it that much(a few try on making it write a js program that can run each 5 minute, that should be able to download a website and check URLs in it using RegEx) it didn't give proper code.

I wonder if you can explain why it behaves like that (I believe data matters in this case, maybe your fine tuning data) or if you can further develop it.

Thanks for this Llm because open source community needs Turkish llms that are robust.

Trendyol org

@JosephH , thanks for your great wishes. We have aimed to create Turkish models using transfer learning. So, we used general knowledge datasets in training. In further iterations, we may also focus on the code generation parts.

yusufcakmak changed discussion status to closed

Sign up or log in to comment