Exploring Text Generation Capabilities in Persian with Specific GPU Requirements

#14
by reyhane12 - opened

Can these quantized models generate text in Persian?
which GPUs are needed to run these models effectively?

Hi,
I am not sure how capable the original Llama-3-70B is in other non-western languages like Arabic or Persian. I think there should be a separate fine-tuned model focusing on Persian language to improve the support and quality.

The model in 16bit requires ~140G VRAM to be loaded into multiple GPUs. (at least 2 A100 with 80G each, or 4 A100 with 40G each)

MaziyarPanahi changed discussion status to closed

Sign up or log in to comment