Text Generation
Transformers
Safetensors
qwen2
reranker
conversational
text-generation-inference

reduced size

#3
by Arman-Sadeghi - opened

Hi there, how did you reduce the size and made it multilingual?
I mean, how can I make it focus on being multilingual(maximum 4 languages) in order to make it smaller and still a nice chatbot?

Lightblue KK. org

Hey Arman,

It is based on a 0.5B model (Qwen/Qwen2.5-0.5B-Instruct) so that is how the size was set.

I am unsure as to how much smaller you could make it by focusing only on a few languages - there is as much of an argument for cross-lingual transfer learning as there is for more focused training.

If you'd like to chat more about this, feel free to PM me on LinkedIn (https://www.linkedin.com/in/peter-devine-030b2012a/) and we can talk more about your problem.

Thanks,

Peter

Sign up or log in to comment