AMDMIGraphX
AMDMIGraphX copied to clipboard
ROCmSoftwarePlatform
AMD's graph optimization engine.
Currently the clang-format and python formatter complains about formatting such that a developer has to manually run the formatting script with the specific version of rocm or do the formatting...
This is the tracking ticket to make sure rocMLIR is shipped by default in MIGraphX. If I understand the conversation with @pfultz2 correctly, this needs mostly infrastructure (docker) related work...
* So the tiling can handle dimensions that are not easily divisible like 1009
* See `src/py/migraphx_py.cpp:155` for the section of code with the current encoding
* Update FP8 OCP support to use hipblaslt when hipblaslt is enabled
Onnx parsers for Quantization Dequantization: case when scales and zero point tensors of same dimension as the input tensor: Currently these parsers try to either broadcast or multibroadcast. Also these...
Tested with driver: 14616 $ rbuild package -d deps -DGPU_TARGETS=gfx1201 ... $ make check -j$(nproc) [ 1%] Built target embed_lib_migraphx_kernels 349/380 Test #378: test_py_3.12_backend ......................................................***Failed 0.24 sec Traceback (most recent...
Checked on MI300X: - [ ] dlrm_criteoterabyte_fp16 - [ ] vicuna_fastchat
