thaisacs opened a new pull request, #18328:
URL: https://github.com/apache/tvm/pull/18328
Before this PR I was getting the following error:
`Traceback (most recent call last):
File "/home/thais/Dev/TVM-LLM-Bench/test-export-gpt.py", line 35, in
<module>
mod: tvm.IRModule = from_exported_program(exported_program)
File
"/home/thais/Dev/new-tvm/tvm/python/tvm/relax/frontend/torch/exported_program_translator.py",
line 806, in from_exported_program
return ExportedProgramImporter().from_exported_program(
File
"/home/thais/Dev/new-tvm/tvm/python/tvm/relax/frontend/torch/exported_program_translator.py",
line 697, in from_exported_program
self.env[node] = self.convert_map[func_name](node)
File
"/home/thais/Dev/new-tvm/tvm/python/tvm/relax/frontend/torch/base_fx_graph_translator.py",
line 409, in convert
return call_binary_op(relax_op, lhs, rhs)
File
"/home/thais/Dev/new-tvm/tvm/python/tvm/relax/frontend/torch/base_fx_graph_translator.py",
line 405, in call_binary_op
return self.block_builder.emit(op(lhs, rhs))
File "/home/thais/Dev/new-tvm/tvm/python/tvm/relax/block_builder.py", line
323, in emit
return _ffi_api.BlockBuilderEmit(self, expr, name_hint) # type: ignore
File "tvm/ffi/cython/./function.pxi", line 228, in
tvm.ffi.core.Function.__call__
File "tvm/ffi/cython/core.cpp", line 24315, in
__pyx_pw_3tvm_3ffi_4core_8Function_1__call__
File "tvm/ffi/cython/core.cpp", line 24369, in
__pyx_pf_3tvm_3ffi_4core_8Function___call__
File "tvm/ffi/cython/core.cpp", line 23995, in
__pyx_f_3tvm_3ffi_4core_FuncCall
File "tvm/ffi/cython/core.cpp", line 23890, in
__pyx_f_3tvm_3ffi_4core_FuncCall3
TypeError: Binary operators must have the same datatype for both operands.
However, R.multiply(lv15, lv18) uses datatype float32 on the LHS (StructInfo of
R.Tensor((14, 14), dtype="float32")), and datatype bool on the RHS (StructInfo
of R.Tensor((14, 14), dtype="bool")).
`
--
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]
