search

quic / aimet

AIMET is a library that provides advanced quantization and compression techniques for trained neural network models.
https://quic.github.io/aimet-pages/index.html
Other
2.08k stars 373 forks source link

tf1.15 perchannel error #2467

Open kia350 opened 11 months ago

kia350 commented 11 months ago

Hi, all when i seted "per_channel_quantization": "True" in the config.json met the error: LookupError: No gradient defined for operation 'conv12_end/ReadVariableOp_quantized' (op type: QcQuantizePerChannel)

conv12_end is the last conv layer in the graph.

how can i fix it? thanks

quic-mangal commented 11 months ago

@kia350, could you provide more details regarding the model, so that we could try this on our end. Thanks

kia350 commented 11 months ago

@quic-mangal
hi, It is a normal U-net model , conv2d / max_pooling and conv2d_transpose. When i don't set "per_channel_quantization": "True" , it run succesfully.