Hello,
First of all, I like to thank you for your research works.
I have my own dataset so I tried to fine tune using it over your pre-trained data result.
When I run main.py with the script below, I had the error message on the title, which means fromage didn't recognize pre-trained checkpoint data file so went to else condition. Although I referred your readme file in micro-detail, I was not able to find any clue to figure out this problem. Maby I miss-understood something. I'm working on A100 under the limited resource condition but A100 is also available too.
Here is my script below for main.py. Please advise me. Thanks.
Hello, First of all, I like to thank you for your research works.
I have my own dataset so I tried to fine tune using it over your pre-trained data result. When I run main.py with the script below, I had the error message on the title, which means fromage didn't recognize pre-trained checkpoint data file so went to else condition. Although I referred your readme file in micro-detail, I was not able to find any clue to figure out this problem. Maby I miss-understood something. I'm working on A100 under the limited resource condition but A100 is also available too. Here is my script below for main.py. Please advise me. Thanks.
python -u main.py \ --dist-url "tcp://127.0.0.1:${randport}" --dist-backend 'nccl' \ --multiprocessing-distributed --world-size 1 --rank 0 \ --dataset= xxx --val-dataset=xxx \ --opt-version='facebook/opt-125m' --visual-model='openai/clip-vit-large-patch14' \ --exp_name='xxxx' --image-dir='/datasets/xxxx' --log-base-dir='runs/' \ --batch-size=4 --val-batch-size=100 --learning-rate=0.0003 --precision='fp32' --print-freq=100\ --resume ='/home/.../pretrained_ckpt.pth.tar'\ --epochs=4\ --dataset_dir='xxxxxx'