AMDMIGraphX
AMDMIGraphX copied to clipboard
ROCmSoftwarePlatform
AMD's graph optimization engine.
The format script we have MIGraphX no longer works with ROCm 6.1 because clang-format is no longer available with ROCm AFAIK. Need to fix the script and what clang-format version...
`migraphx-driver` doesn't seem to quantize to int8 or fp8 https://github.com/ROCm/AMDMIGraphX/blob/1199fbeeb83539a5cb6a033745e67bfb661875ca/src/driver/verify.cpp#L97 Need to add quantization for 8 bits and combination of `--fp16 --int8`.
I have some questions about migraphx quantization. First, can migraphx support per-channel quantization of weights ?because per-channel quantization of weights will have better accuracy. Second,can migraphx support FP6+INT8?which has better...
Avoid the overflow in calculations of variance..
ROCm 6.1 was released today. Bumping CI up
NegativeLogLikelihoodLoss is an OP that's similar to that of SoftmaxCrossEntropyLoss described in #2976 Create this function then leverage it for use with SoftmaxCrossEntropyLoss
Torch benchmarks fail. Affected models: Torchbench: speech_transformer
Torch benchmarks fail. Affected models: Torchbench: Background_Matting
Different SKUs have different number of compute units. It is highly desirable to fully utilize the compute hardware, by ensuring that each compute unit is occupied during the execution of...
