Text/vision parameter split
#3
by
AlexThompson
- opened
what are the parameters split of vision part versus text part
We use clip encoder for the vision part of around 400M params and 0.5B params for the text part, which is Qwen-2.5-0.5B