RuntimeError: CUDA error: no kernel image is available for execution on the device

[james20141606]

Hi, thanks for the codes!
I run into a problem when executing:

import torch
from performer_pytorch import Performer

model = Performer(
    dim = 512,
    depth = 1,
    heads = 8,
    causal = True
).cuda()

x = torch.randn(1, 2048, 512).cuda()
y = model(x) # (1, 2048, 512)
x.shape, y.shape

It works fine if I don't use cuda, but when I use cuda it said:

---

RuntimeError Traceback (most recent call last)
in
10
11 x = torch.randn(1, 2048, 512).cuda()
---> 12 y = model(x) # (1, 2048, 512)
13 x.shape, y.shape

/ext3/miniconda3/lib/python3.8/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
720 result = self._slow_forward(*input, **kwargs)
721 else:
--> 722 result = self.forward(*input, **kwargs)
723 for hook in itertools.chain(
724 _global_forward_hooks.values(),

/ext3/miniconda3/lib/python3.8/site-packages/performer_pytorch/performer_pytorch.py in forward(self, x, **kwargs)
341
342 def forward(self, x, **kwargs):
--> 343 return self.net(x, **kwargs)
344
345 class PerformerLM(nn.Module):

/ext3/miniconda3/lib/python3.8/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
720 result = self._slow_forward(*input, **kwargs)
721 else:
--> 722 result = self.forward(*input, **kwargs)
723 for hook in itertools.chain(
724 _global_forward_hooks.values(),

/ext3/miniconda3/lib/python3.8/site-packages/performer_pytorch/reversible.py in forward(self, x, **kwargs)
136
137 for (f, g), (f_args, g_args) in layers_and_args:
--> 138 x = x + f(x, **f_args)
139 x = x + g(x, **g_args)
140 return x

/ext3/miniconda3/lib/python3.8/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
720 result = self._slow_forward(*input, **kwargs)
721 else:
--> 722 result = self.forward(*input, **kwargs)
723 for hook in itertools.chain(
724 _global_forward_hooks.values(),

/ext3/miniconda3/lib/python3.8/site-packages/performer_pytorch/performer_pytorch.py in forward(self, x, **kwargs)
215 self.fn = fn
216 def forward(self, x, **kwargs):
--> 217 return self.fn(self.norm(x), **kwargs)
218
219 class Chunk(nn.Module):

/ext3/miniconda3/lib/python3.8/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
720 result = self._slow_forward(*input, **kwargs)
721 else:
--> 722 result = self.forward(*input, **kwargs)
723 for hook in itertools.chain(
724 _global_forward_hooks.values(),

/ext3/miniconda3/lib/python3.8/site-packages/performer_pytorch/performer_pytorch.py in forward(self, x, context, mask, context_mask)
297 attn_outs.append(out)
298
--> 299 out = torch.cat(attn_outs, dim = 1)
300 out = rearrange(out, 'b h n d -> b n (h d)')
301 out = self.to_out(out)

RuntimeError: CUDA error: no kernel image is available for execution on the device

[james20141606]

I checked that I could run some simple model with cuda. I am not sure if performer requires some certain environment. I am using pytorch==1.6.0 with cuda 10.1