Ziheng Wang | Sree Harsha Nelaturu
The repository for the workshop paper presented at EMC^2, ISCA 2019.
To reproduce results for CIFAR10:
- Ensure you have CUDA 8 and Tensorflow GPU Installed (Tested on TF 1.9)
- In the CUDA_Kernels/ folder, run create_ops.sh.
- Place the compiled approx_kernel2.so in the model_harness/models/
- Run
python cifar10_epoch_wise.py