dlexplorer opened a new issue #7096:
URL: https://github.com/apache/tvm/issues/7096
Many frameworks define dynamism in the shape as -1 in the certain axis while
TVM has its own constant for this - relay.Any().
If I take onnx bert model and try to convert like below snippet code, TVM
will crash.
Two issues
1. not crash
2. During conversion need to identify -1 in the shape and warn that TVM uses
its own id for defining of dynamic shape/stop execution
Probably stopping of execution after determination of -1 will solve the
crash as well
snippet:
```
import argparse
import sys
import onnx
import tvm
from tvm import relay
parser = argparse.ArgumentParser(description=
"Converts and compiles ONNX model")
required = parser.add_argument_group('required arguments')
required.add_argument('-m', '--input_model', required=True, type=str,
help="path to ONNX model")
args = parser.parse_args()
onnx_model = onnx.load(args.input_model)
shape_dict = {}
shape_dict["input_mask:0"] = [-1,256]
shape_dict["segment_ids:0"] = [-1,256]
shape_dict["input_ids:0"] = [-1,256]
shape_dict["unique_ids_raw_output___9:0"] = [-1,256]
mod, params = relay.frontend.from_onnx(onnx_model, shape_dict)
target = "llvm"
with tvm.transform.PassContext(opt_level=3):
lib = relay.build(mod, target=target, target_host=target, params=params)
lib.export_library(args.input_model + ".so")
```
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
