Thrsu opened a new issue, #15758:
URL: https://github.com/apache/tvm/issues/15758
I encountered a KeyError while trying to convert a Torch FX model with a
layer normalization operation to TVM using the from_fx function from the
tvm.relax.frontend.torch module.
### Actual behavior
```
Traceback (most recent call last):
...
mod = from_fx(fx_model, input_info)
File
"/workplace/software/tvm/tvm/python/tvm/relax/frontend/torch/fx_translator.py",
line 1468, in from_fx
return TorchFXImporter().from_fx(
File
"/workplace/software/tvm/tvm/python/tvm/relax/frontend/torch/fx_translator.py",
line 1355, in from_fx
self.env[node] = self.convert_map[func_name](node)
File
"/workplace/software/tvm/tvm/python/tvm/relax/frontend/torch/fx_translator.py",
line 877, in _layer_norm
gamma = self.env[node.kwargs["weight"]]
KeyError: None
```
### Environment
- branch: unity
- tvm: v0.14.dev0-595-gfa0d35b
- torch: 2.0.1
### Steps to reproduce
```python
import torch
from torch import fx
from torch.nn import Module
import tvm
from tvm import relax
from tvm.relax.frontend.torch import from_fx
input_data = torch.randn([4, 2, 2, 5], dtype=torch.float32)
para_1 = (2, 2, 5)
para_2 = None
para_3 = None
para_4 = 0.001
class layer_norm(Module):
def forward(self, input):
return torch.nn.functional.layer_norm(input,
para_1,para_2,para_3,para_4,)
model = layer_norm().float()
input_info = list(zip([list(inp.shape) for inp in input_data],
[str(inp.dtype) for inp in input_data]))
fx_model : torch.fx.GraphModule = fx.symbolic_trace(model)
with torch.no_grad():
mod = from_fx(fx_model, input_info)
```
### Triage
* needs-triage
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
