intel-extension-for-pytorch
intel-extension-for-pytorch copied to clipboard
intel
A Python package for extending the official PyTorch that can easily obtain performance on Intel platform
I was playing around with a debug model with some convolutions, but got the input shapes of some intermediate layer wrong (in my case a Conv after a Linear, i.e....
I am trying to set up a function to use IPEX and call the inference with the minimum load time as possible That means I want to reuse the traced_model...
Observations: - Using fused optim (e.g. Adam) + `ipex.optimize(..., level='01', dtype=torch.bfloat16)` **no error** - non fused optimizer (e.g. AdamW) + `ipex optimize(..., level='01', dtype=torch.float32)` + `AMP bf16` **no error** -...
We got an "Illegal instruction (core dumped)" error when using IPEX 1.12.0 in machines without AVX-512. It used to work on 1.11.0.
I am using version 1.9.0 ``` Traceback (most recent call last): File "./timing.py", line 106, in ipex.optimize(model.backbone) File "/data/home/ubuntu/projects/ai-general/ml/.venv/lib/python3.8/site-packages/intel_extension_for_pytorch/frontend.py", line 238, in optimize optimized_model = optimization.fuse(optimized_model, inplace=inplace) File "/data/home/ubuntu/projects/ai-general/ml/.venv/lib/python3.8/site-packages/torch/fx/experimental/optimization.py", line...
Hi, I am Nathan. These day I tried to make inference based on libtorch(c++) speed up by using intel-extension I am using wenet-e2e toolkit for speech recognition system. As follow...
Hi, I am trying to compile this extension using debian 11. The compilation went well but in the linking stage I got the following error: ``` ... [100%] Linking CXX...
The script at "[test_weight_prepack.py](https://github.com/intel/intel-extension-for-pytorch/blob/master/tests/cpu/test_weight_prepack.py)" has some errors as follows: - ipex.optimize has no attribute as 'sample_input' : `ipex.optimize(origin_model1, dtype=dtype, optimizer=origin_optimizer1, level='O1', sample_input=x)` - Issues with 3d torch tensor for NWC...
1)Vertical split embedding to scale-out to much more ranks. 2)LAMB to enable large batch size.
