AgitatedDove14 Thanks for advise, I think it would work for tflite. However, SNPE performs quantization with precompiled CLI binary instead of python library (which also needs to be installed). What would be the pipeline in this case?
CostlyOstrich36 Typical workflow is the following:
Train object detection CNN Pick the best checkpoint Perform https://www.tensorflow.org/lite/performance/post_training_quantization in tflite, OpenVino or SNPE to obtain quantized model. It requires a set of representative image files.What I'd like to do is to run quantization procedure for various frameworks (listed above) for a given task on its' completion
AgitatedDove14 , one more question: in case of triggering the quantization process, will it be considered as separate task? I mean, will I be able to clone it and rerun with different parameters (bitness, optimizations, etc) for the same trained network?
MistakenBee55 how about a Task doing the Model quantization, then trigger it with TriggerScheduler ?
https://github.com/allegroai/clearml/blob/master/examples/scheduler/trigger_example.py
However, SNPE performs quantization with precompiled CLI binary instead of python library (which also needs to be installed). What would be the pipeline in this case?
I would imagine a container with preinstalled SNPE compiler / quantizer, and a python script triggering the process ?
one more question: in case of triggering the quantization process, will it be considered as separate task?
I think this makes sense, since you probably want a container with the SNE environment, meaning you cannot use a simple trigger function. wdyt?