Ma Mingfei

@jcjohnson @renato2099 we have optimized torch performance on CPU platform. CPU optimization is provided by mklnn and mkltorch, to install torch with mklnn and mkltorch, refer to https://github.com/intel/torch currently we...

This patch enables `channels last` support for THNN's native transposed conv2d implementation, to ensure `convTranspose2d` will have identical behavior with v.s. without mkldnn on the CPU side.

@VitalyFedyunin, both #70897 and #74023 have dependency on #77060 (this was reverted due to test case failures, i have fixed it but it requires upgrade of `ideep`, so i marked...

Reland #58348. **Need to upgrade ideep first to modify the memory format propagation logic.** put it under draft at the moment. Original error from https://hud.pytorch.org/pytorch/pytorch/commit/479e0d64e607d48fc61a958caf6aeaf165e3d45d is due to misjudge a...

> Could you clear the OSS CI? @mingfeima The thing is our oneDNN-to-pytorch integration logic is mostly implemented in 'ideep' which is a third party repo to pytorch. We will...

Update: old failures fixed by ideep upgrade. also lifting restrictions from `torch/testing/_internal/common_modules.py` since `ConvTranspose2d` has channels last support now, otherwise `python test_modules.py TestModuleCPU.test_memory_format_nn_ConvTranspose2d_cpu_float32` would fail.

@xuzhao9 I will have the regression fixed. Which command should I use if I want to run a standalone test, for example this one `hf_BigBird-cpu-eager` ?

@malfet, I am still working on this and I will put detailed explanations when having the issue fixed.

@xuzhao9 sorry I haven't fixed this issue so far, I am still experimenting.

@xuzhao9 do you have `profiler` log with and without https://github.com/pytorch/pytorch/commit/4e9b969baa6 on `python3 run.py -t train -m eager hf_BigBird -d cpu` ? If you can get profiler log on your side,...