brevitas

brevitas copied to clipboard

Add support for PReLU.

volcacius

Zero biases quantized to NaN

3

I'm attempting to implement a quantized version of pytorch's [LayerNorm](https://pytorch.org/docs/stable/generated/torch.nn.LayerNorm.html). My implementation is functional when weights and biases are disabled and it is purely acting functionally. However, the biases in...

tafk7

Input scale required

9

Hi, I am running the example in brevitas github page as follows. I am experiencing the following issue: ```python Exception has occurred: RuntimeError Input scale required File "/home/brevitas/proxy/parameter_quant.py", line 196,...

ardeal

Hi, For the following code, we defined qnn.QuantIdentity with input parameter bit_width=4, but in there is not bit_width in QuantIdentity class. My questions are: how will the bit_width=4 passed to...

ardeal

output of QuantReLU is bigger than 8

1

Hi, For the following code, max_val is set to 8.0, however, I could see that the output of relu1 may be bigger than 8. why? ```python self.relu1 = qnn.QuantReLU(quant_type=QuantType.INT, bit_width=16,...

ardeal

Hi, i'm tryin to run the validation of MobileNet V1. When I run: python imagenet_val.py --pretrained I have the following error: raceback (most recent call last): File "/home/nurbano/Documentos/brevitas/src/brevitas_examples/imagenet_classification/imagenet_val.py", line 128,...

nurbano

Hi, thanks for your great work, I have a question about the QuantTanh, maybe it's a pytorch problem when I use the QuantTanh, fater I saved the model, I cannot...

tricky61

Improvement: Change review_requested trigger to issue_comment trigger for full test suite

7

## Reason for this PR https://github.com/Xilinx/brevitas/issues/869 Full test suite is currently run when a review is requested which means that if several people were selected as reviewers it would trigger...

costigt-dev

## Issue Implements https://github.com/Xilinx/brevitas/issues/727 ## Details Creates test file for QuantTensor Has tests for: - maths operators - divide by zero - transpose - view - is_valid It's not exhaustive...

costigt-dev

Quant layer should not expose any attribute of their underlying proxys (e.g, quant_output_scale in QuantLayer). Similarly, caching should be delegated to Proxy. Missing tasks: - [x] Remove comments - [x]...

Giuseppe5