SparseML quantization

[ghost]

[bfineran]

Hi @TomTels, input and sections are submodules here in the resnet50 model that this recipe was applied to. They can be changed to names of submodules in a different model to be quantized.

[ghost]

thnak you for your comment!, I actually replaced thses names with modules of Resnet50 such as

  submodules:
     - conv1
     - layer1
     - layer2
     - layer3
     - layer4
     - fc

but, the quantization seems to fail. benchmark_result = engine.benchmark() is too slow.

• default: ms_per_batch: 9.87
• pruning85: ms_per_batch: 5.730035305023193
• pruning88: ms_per_batch: 5.463829040527344
• qunantization: 61.941800117492676

(I actully posted the same question in Slack:Deep Sparse Community because I though that familiar people join there, if my question is duplicated here, pls remove my question here.)