error: 'ttnn.max_pool2d' op ttnn.max_pool2d currently only supports an input type of bfloat16. Recieved 'f32'.

[kamalrajkannan78]

Description

• error: 'ttnn.max_pool2d' op ttnn.max_pool2d currently only supports an input type of bfloat16. Recieved 'f32'.

Reproduce

git checkout kkannan/models_mlir_issues 
git submodule update --recursive
cmake --build build -- install_ttforge
pytest forge/test/model_demos/high_prio/cnn/pytorch/test_vovnet.py -k "not timm" -svv

Observed Behaviour

error: 'ttnn.max_pool2d' op ttnn.max_pool2d currently only supports an input type of bfloat16. Recieved 'f32'.

Logs

• vovnet_issue.log

TTIR graphs

• vovnet27s_ttir.txt
• vovnet39_stigma_ttir.txt
• vovnet39_ttir.txt
• vovnet57_stigma_ttir.txt
• vovnet57_ttir.txt

[meenakshiramanathan1]

More details in tenstorrent/tt-metal#15768

[sdjordjevicTT]

We fixed temporarily this on the MLIR side with the workaround that casts input to bf16 introduced in:
#1657

@meenakshiramanathan1 @kamalrajkannan78 can you confirm that you don't see this failure anymore?

[sdjordjevicTT]

I will close this one, if you find it fails again, pleas reopen the issue.