search

etrommer / torch-approx

GPU-accelerated Neural Network layers using Approximate Multiplications for PyTorch
https://etrommer.de/torch-approx
MIT License
6 stars 3 forks source link

Benchmark against TFApprox #7

Open etrommer opened 1 year ago

etrommer commented 1 year ago

Some comparison with TFApprox with TFApprox was requested.

This should be kept separate from productive code, similar to #6

Test case is not fully-defined yet. Most likely scenario: Comparison of Conv2D inference speed.

etrommer commented 1 year ago

Build Environment used for TorchApprox: CUDA 10.1 Tensorflow 2.3.0 CuDNN 7.6.5

CUDA versions >= 11.0 do not compile due to incompatible API

etrommer commented 1 year ago

Rerun after merging #8

conv_layer_bench.zip

etrommer commented 1 year ago

Rerun using all High-Througput Models

conv_layer_benchmark.zip