RuntimeError: "slow_conv2d_cpu" not implemented for 'Half'

[InugYoon]

Hello, I followed the cocoop project, and occured the error above.

My progress was as follows.

1. Env setup: following setups from dassl page and cocoop page
2. Dataset and json file download
3. Folder setup: First I git cloned the coop project, then copied dassl folder which was downloaded by git cloning dassl project.
4. Script: bash script/cocoop/base2new_train.sh caltech101 2

Then I get the following error.
Initialize tensorboard (log_dir=output/base2new/train_base/caltech101/shots_16/CoCoOp/vit_b16_c4_ep10_batch1_ctxv1/seed2/tensorboard)
Traceback (most recent call last):
File "train.py", line 207, in
main(args)
File "train.py", line 150, in main
trainer.train()
File "/root/CoOp/dassl/engine/trainer.py", line 386, in train
super().train(self.start_epoch, self.max_epoch)
File "/root/CoOp/dassl/engine/trainer.py", line 250, in train
self.run_epoch()
File "/root/CoOp/dassl/engine/trainer.py", line 596, in run_epoch
loss_summary = self.forward_backward(batch)
File "/root/CoOp/trainers/cocoop.py", line 264, in forward_backward
loss = model(image, label)
File "/root/anaconda3/envs/coop_test/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "/root/CoOp/trainers/cocoop.py", line 178, in forward
image_features = self.image_encoder(image.type(self.dtype))
File "/root/anaconda3/envs/coop_test/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(args, **kwargs)
File "/root/CoOp/clip/model.py", line 220, in forward
x = self.conv1(x) # shape = [, width, grid, grid]
File "/root/anaconda3/envs/coop_test/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "/root/anaconda3/envs/coop_test/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 463, in forward
return self._conv_forward(input, self.weight, self.bias)
File "/root/anaconda3/envs/coop_test/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 459, in _conv_forward
return F.conv2d(input, weight, bias, self.stride,
RuntimeError: "slow_conv2d_cpu" not implemented for 'Half'

When I google it, only results show the 'whisper' library issue which is from openai.
Do you have any idea about this issue?

[AtsuMiyai]

@InugYoon
I consider that torch cannot identify GPUs and uses CPUs. Could you check the torch version of your environment?

[Ezio77]

@InugYoon
I met this error before, and finally figured out that i have installed pytorch with a wrong version. Make sure pytorch version fit your cuda version so that torch.cuda.is_available() is True.

[joy-lewis]

@InugYoon I'm working on a M2 Mac and I solved this issue by changing the line cfg.TRAINER.COOP.PREC = "fp16" # fp16, fp32, amp to cfg.TRAINER.COOP.PREC = "fp32" # fp16, fp32, amp , inside the CoOp/train.py file.