/mnt/petrelfs/leihaodong/anaconda3/envs/eval_xx/lib/python3.10/site-packages/torchmetrics/utilities/prints.py:43: UserWarning: Metric `InceptionScore` will save all extracted features in buffer. For large datasets this may lead to large memory footprint. warnings.warn(*args, **kwargs) /mnt/petrelfs/leihaodong/anaconda3/envs/eval_xx/lib/python3.10/site-packages/open_clip/factory.py:450: UserWarning: QuickGELU mismatch between final model config (quick_gelu=False) and pretrained tag 'openai' (quick_gelu=True). warnings.warn( Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.52, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`. Loading PartiPrompts from /mnt/petrelfs/leihaodong/ICLR25/sjdtree/data/prompts/PartiPrompts.tsv for GPU 0, Node 0 Number of unique prompts: 1632 | Max prompt length: 389 一共932 张 Evaluating images: 0%| | 0/932 [00:00