Qwen3.5 trained on a wide variety of available datasets ?

#1
by Rebis - opened

Hi,
Do you plan to create a version of Qwen3.5 trained on a wide variety of available datasets, as is the case for this model ?
If so, it would be interesting to train these multimodal capabilities as well, or at least retain them.
Thank you for everything.

Rebis changed discussion title from Qwen 3.5 trained on a wide variety of available datasets ? to Qwen3.5 trained on a wide variety of available datasets ?
Owner

Hi Rebis,

Thank you for your attention to my work!

I am a simple hobbist and currently can only afford free kaggle GPU. My primary goal to finetune Qwen3-4B-Instruct is to enhance its capability on low VRAM personal devices while not generating lengthy chain-of-thought.

I suppose Qwen3.5-4B is already powerful enough without thinking too much even without any finetune. If I am still to finetune it, I will focus on Instruct mode rather than CoT or multi-modal.

Thank you for your understanding!

Sign up or log in to comment