The problem of fine-flickr30k

dandelin / ViLT

Code for the ICML 2021 (long talk) paper: "ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision"

Apache License 2.0

1.41k stars 208 forks source link

The problem of fine-flickr30k #75

Open wuqiang12345 opened 2 years ago

wuqiang12345 commented 2 years ago

Hello, what configuration does your vilt use when fine-tuning flickr30k? Eight gpus? What is the memory of each gpu? What is the result of fine-tuning flickr30k? Is it a weight file?