knjwhn edited a comment on issue #16749: Ask for advice about using my int8gemm URL: https://github.com/apache/incubator-mxnet/issues/16749#issuecomment-558002487 > Are you running experiments or targeting for upstreaming in the future? If just experiments, I think you can directly replace the `cblas_gemm_s8u8s32` function call with yours. Before that, you need make sure that they have the same functionality. See the document of `cblas_gemm_s8u8s32` at https://software.intel.com/en-us/mkl-developer-reference-c-cblas-gemm-1 > > If you're not building MXNet with `USE_BLAS=mkl`, remember to remove the check of >`MSHADOW_USE_MKL == 1` around that function. Also avoid using things like `MKL_INT8`. @TaoLv hello, I'm so sorry if I bother you . and I have a question here. I directly replace the `cblas_gemm_s8u8s32` function with my function, and I got the similar result when I am running example/quantization/ with the command from quantization/README.md `python imagenet_inference.py --symbol-file=./model/mobilenet1.0-quantized-5batches-naive-symbol.json --param-file=./model/mobilenet1.0-quantized-0000.params --rgb-mean=123.68,116.779,103.939 --rgb-std=58.393,57.12,57.375 --num-skipped-batches=50 --batch-size=64 --num-inference-batches=500 --dataset=./data/val_256_q90.rec --ctx=cpu` and the result I got is here : before I replace: INFO:logger:Finished inference with 32000 images INFO:logger:Finished with 24.355488 images per second INFO:logger:('accuracy', 0.72) INFO:logger:('top_k_accuracy_5', 0.90475) after I replace: INFO:logger:Finished inference with 32000 images INFO:logger:Finished with 23.792868 images per second INFO:logger:('accuracy', 0.7173125) INFO:logger:('top_k_accuracy_5', 0.9046875) so, I was wondering if the code really called the QuantizedFullyConnectedForwardCPU function which contains the cblas_gemm_s8u8s32 function, and i add some print message but not shows finally. and I want to know how can I use the function called QuantizedFullyConnectedForwardCPU , Hope for your help.
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
