You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Is your feature request related to a problem? Please describe.
I am currently trying to run segment anything in the browser in a rust binary that is targeting wasm. I am using the vit_b onnx model provided in the list here (direct download)
It took many iterations to get the model into the state where i hit this issue.
Removed all dynamically sized inputs and intermediaries
Ran through onnx-simplifier to remove some of the operations that nnx was not able to infer the shape of
Hacked in support for inference on Not and Equal operators (some likelihood this was not done correctly, but should not be the source of the main issue)
Removed the dynamically sized outputs when running nnx prepare on the model
At this point when running the model I was faced with a IrError(Type(NotSupported(BOOL)) which indicates the model is expecting a data type that does not fit into the existing supported list.
Each of these workarounds does pose a serious limitation to use wonnx for this task. But it doesn't seem this final one can be worked around.
It is possible that after this, the model is still unusable for other reasons.
Describe the solution you'd like
Implementation of BOOL data type support
Describe alternatives you've considered
Restructuring the model in various ways to fit current support.
Additional context
While this issue is strictly looking at the BOOL data type, it would be awesome to have the other shortcomings mentioned looked at as well.
The text was updated successfully, but these errors were encountered:
Is your feature request related to a problem? Please describe.
I am currently trying to run segment anything in the browser in a rust binary that is targeting wasm. I am using the vit_b onnx model provided in the list here (direct download)
It took many iterations to get the model into the state where i hit this issue.
onnx-simplifier
to remove some of the operations that nnx was not able to infer the shape ofNot
andEqual
operators (some likelihood this was not done correctly, but should not be the source of the main issue)nnx prepare
on the modelAt this point when running the model I was faced with a
IrError(Type(NotSupported(BOOL))
which indicates the model is expecting a data type that does not fit into the existing supported list.Each of these workarounds does pose a serious limitation to use
wonnx
for this task. But it doesn't seem this final one can be worked around.It is possible that after this, the model is still unusable for other reasons.
Describe the solution you'd like
Implementation of
BOOL
data type supportDescribe alternatives you've considered
Restructuring the model in various ways to fit current support.
Additional context
While this issue is strictly looking at the
BOOL
data type, it would be awesome to have the other shortcomings mentioned looked at as well.The text was updated successfully, but these errors were encountered: