Bhavya Medishetty
Bhavya Medishetty
This PR is dependent on https://github.com/pytorch/pytorch/pull/74704 Remove the redundant call for hipify from setup.py - build -passes - pytest test/test_transforms.py -- ``` SKIPPED [1] test/test_transforms.py:156: accimage not available SKIPPED [1]...
This is a temporary fix for the following error , which is encountered when running [stable_diffusion inference with deepspeed inference ](https://github.com/philschmid/transformers-deepspeed/blob/main/inference/stable_diffusion_optimization.ipynb) till __double2half intrinsic support is enabled on rocm. ```[1/9]...
I got this error when trying to run stable_diffusion inference with deepspeed TypeError: dot() got an unexpected keyword argument 'trans_b' Issue found on triton with the same error : [TypeError:...
[triton flash_attention](https://github.com/microsoft/DeepSpeed/blob/master/deepspeed/ops/transformer/inference/triton_ops.py) used by deepspeed inference is compatible with 6 month old pre release version of triton( 2.0.0.dev20221202 ) . triton dot , trans operators are rewritten in [Complete rewrite...
Update deepspeed/ops/transformer/inference/triton_ops.py with latest triton/python/tutorials/06-fused-attention.py, num_stages = 1 in deepspeed/ops/transformer/inference/triton_ops.py , num_stages=2 in triton/python/tutorials/06-fused-attention.py, because when running stable diffusion inference with deepspeed inference engine with num_stages=2 gives out of memory...
Changes to add hipify_torch (hipification for amd-build) as submodule Build - passes build/gloo/test/gloo_test - passes This is a companion PR to https://github.com/pytorch/pytorch/pull/74704, which adds hipify_torch submodule to pytorch
This PR is to enable tests which are skip on rocm from test package test_unary_ufuncs.py::TestUnaryUfuncsCUDA test_file | test_name | test_class -- | -- | -- test_unary_ufuncs | test_reference_numerics_large_polygamma_polygamma_n_2_cuda_float16 | (__main__.TestUnaryUfuncsCUDA)...
### 🚀 The feature, motivation and pitch I am working on enabling test_nn.py test_cudnn_weight_format on rocm, an observed that the test works if ``` diff --git a/test/test_nn.py b/test/test_nn.py index c8311c91d7..85b391e880...