Open rezarokni opened 3 years ago
The use of tft.layer within ProcessReturn class in the library is inefficient for batch inference. The output of this layer can be made available via the serving signature which avoids double instantiation and processing in the pipeline.
Correction being worked on in branch : TS-36-Remove-tftlayer-post-process
The use of tft.layer within ProcessReturn class in the library is inefficient for batch inference. The output of this layer can be made available via the serving signature which avoids double instantiation and processing in the pipeline.