cudnn-frontend
cudnn-frontend copied to clipboard
NVIDIA
cudnn_frontend provides a c++ wrapper for the cudnn backend API and samples on how to use it
I am using the cudnn_frontend to perform a simple matmul - all works as expected when using FLOAT data type, but the graph::check_support is really slow. I there any way...
Is this [link](https://github.com/NVIDIA/cudnn-frontend/blob/9f82dda5c029d15a5f371f0fe003dc0c74a0c987/samples/legacy_samples/f16_flash_mha_sample.cpp#L714C14-L714C14) a flash attention?
When I configure with `-DCUDNN_FRONTEND_BUILD_UNIT_TESTS=ON`, I get ``` CMake Error at CMakeLists.txt:49 (add_subdirectory): The source directory /path/to/cudnn-frontend/1.1.0/test does not contain a CMakeLists.txt file. ``` Is the `test/` directory supposed to...
include\cudnn_frontend\graph_interface.h(444,19): Error C2248: 'cudnn_frontend::graph::Layernorm_attributes::forward_phase': cannot access private member declared in class 'cudnn_frontend::graph::Layernorm_attributes'
When importing cudnn-frontend as a 3rd party library using FetchContent, cmake incorrectly parse the file path. This is because we incorrectly using CMAKE_SOURCE_DIR variable, which should be PROJECT_SOURCE_DIR (The former...
How could I disable channel-last memory layout in the graph.conv_fprop(...) function ? So the layout is NCHW instead. Thanks.
Hello, I've been reading through the documentation, especially regarding the convolution operation and I'm confused on what is the default mode for the convolution operation, whether it is `CUDNN_CONVOLUTION` or...
I believe I've found a bug within the Flash with rng dropout sample test ``` if (seq_len_override) { Surface devActualSeqlenQ(b, false); Surface devActualSeqlenKV(b, false); std::vector hostActualSeqlenQ(b, 20); std::vector hostActualSeqlenKV(b, 20);...
I encountered a test failure after building and running the tests. Here are the details: - **GPU**: RTX 4090 - **Repo branch**: v1.4.0 - **Operating System**: Ubuntu 22.04.3 - **CUDA...
when running `pytest test/python_fe` on latest version, it returns ``` graph.validate() graph.build_operation_graph() graph.create_execution_plans([cudnn.heur_mode.A, cudnn.heur_mode.FALLBACK]) > graph.check_support() E cudnn._compiled_module.cudnnGraphNotSupportedError: [cudnn_frontend] Error: No execution plans built successfully. test/python_fe/test_matmul_bias_relu.py:278: cudnnGraphNotSupportedError ==================================================================================== warnings summary...
