Fail to convert the bie to nef due to invalid bit width
Dear sir,
We have faced a conversion issue stated by "Invalid program input: output bit width [16] of node [Conv_414]"
Could you please help us to resolve this issue? thanks!
See the attachment to find the onnx (parsed by onnx2onnx_flow, optimized.7z) and the conversion script (test.zip).
Tagged:
The discussion has been closed due to inactivity. To continue with the topic, please feel free to post a new discussion.
Comments
I just found this message if the number of the groups in grouped convolution is larger than 40 or 60.
Is it resolvable or not? thanks.
Hi Karasu,
Sorry for the delay; we're currently looking into the issue with another team, and we'll let you know when we've found out the cause for this issue.
Thank you for your patience!
@Maria Chen
Many thanks for your reply.
A further question is: Could we have a non-image format as the input? I just modified the kp_core.c and kp_inference.c for allowing the different data types but it seems to not work. Just wonder that.
@Karasu Hsu
Hi Karasu,
Could you provide a model with less than 40 or 60 group convolutions for us to investigate further?
Sure, please find the #groups=40 version in the
theaattachment.
Hi Karasu,
Yes, we could have a non-image format as the input.
If you were using Kneron PLUS v1.3.0, you could use the function kp_generic_raw_inference_bypass_pre_proc_send to bypass the pre-processing and inference your input. You would also need to be aware of the NPU Data format (please refer to the link: http://doc.kneron.com/docs/#plus_c/appendix/supported_npu_data_layout_format/) when you load the non-image input.
Reference links: Inference API - Document Center (kneron.com)
Migrate to Kneron PLUS 2 - Document Center
If you were using Kneron PLUS v2.0.0, you could also use the function kp_generic_data_inference_send to inference a non-image input.
Reference link: Inference API - Document Center (kneron.com)
Many thanks!