Why the tensor allocation strategy is less when unet model using static shape? #16240
Unanswered
KotomiHacker
asked this question in
Other Q&A
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
I did an experiment of running unet model with static shape and dynamic shape and printing the tensor allocation strategy o each. The results are totally different.
For static shape, the comnmand line is
onnxruntime_perf_test.exe -m times -r 1 -f unet_sample_batch:2 -f unet_sample_channels:4 -f unet_sample_width:64 -f unet_sample_height:64 -f unet_hidden_batch:2 -f unet_hidden_sequence:77 -e dml olive_sd1.5\unet\model.onnx
. I only get four ortValue as below with static shape while there are more tensors about allocation information with dynamic_shape.Beta Was this translation helpful? Give feedback.
All reactions