tinywisdom opened a new issue, #18339:
URL: https://github.com/apache/tvm/issues/18339
### Expected behavior
+ The frontend should support lowering `torch.mm` into an equivalent Relax
operator (likely `relax.op.matmul`).
+ At minimum, if unsupported, provide a clearer user-facing error message
and possible fallback guidance.
### Actual behavior
When converting a PyTorch `torch.exported` program into TVM Relax using
`from_exported_program`, a model containing `torch.mm` fails with:
```
AssertionError: Unsupported function types ['mm.default']
```
This indicates that the `mm.default` operator (2D matrix multiply) is
currently not supported in the TVM Relax PyTorch frontend.
### Environment
+ OS: (Ubuntu 22.04.4 LTS (x86_64))
+ TVM version: (release v0.21.0)
+ Python: (3.10.16)
+ LLVM: (17.0.6)
### Steps to reproduce
```python
import torch
import torch.nn as nn
from torch.export import export as torch_export
from tvm.relax.frontend.torch import from_exported_program
class M(nn.Module):
def forward(self, a, b):
# Key op: torch.mm (2D matrix multiply)
return torch.mm(a, b)
def main():
torch.manual_seed(0)
m = M().eval()
# Inputs: (2, 3) @ (3, 4) -> (2, 4)
a = torch.randn(2, 3, dtype=torch.float32)
b = torch.randn(3, 4, dtype=torch.float32)
# 1) Check eager path
with torch.inference_mode():
y = m(a, b)
print("PyTorch eager OK, y.shape =", tuple(y.shape))
# 2) Export
ep = torch_export(m, (a, b))
print("ExportedProgram created.")
# 3) Import into TVM Relax — triggers unsupported function type
mod = from_exported_program(ep)
if __name__ == "__main__":
main()
```
### Triage
* needs-triage
* bug
--
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]
