Describe the issue
I am currently trying to add support for AMD GPUs in an onnxruntime-based project. I am using the ort wrapper for Rust. For some reason model loading fails. There are three initial spikes on the CPU followed by small ones on the GPU. Then one core keeps being maxed out at almost 100%. Nothing happens for two hours then it fails.
It works well on the CPU and CUDA. I am pretty new to machine learning. Here is the call
https://github.com/jclsn/stemgen/blob/c41cb010fa8bbd458723623264305e66fb87569d/src/demucs.rs#L126
and the log
❯ target/release/stemgen generate --device migraphx ~/Musik/some_track.mp3 ~/Musik/stem-out --model
htdemucs.onnx
[DEMUCS] Creating new Demucs instance with device: migraphx
[DEMUCS] Attempting to use MIGraphX execution provider
[DEMUCS] MIGraphX execution provider configured
[DEMUCS] ORT initialization took: 19.862µs
[DEMUCS] Loading model from local path: "htdemucs.onnx"
Error rewrite_reshapes: /usr/src/debug/migraphx/AMDMIGraphX-rocm-7.2.2/src/include/migraphx/check_shapes.hpp:220: same_dims:
pointwise: Dimensions do not match
Dump: "/tmp/migraphx/rewrite_reshapes56920015489305.mxr"
Error fuse_reduce: /usr/src/debug/migraphx/AMDMIGraphX-rocm-7.2.2/src/include/migraphx/check_shapes.hpp:220: same_dims: poin
twise: Dimensions do not match
Dump: "/tmp/migraphx/fuse_reduce56973354655245.mxr"
Error fuse_pointwise_reduce: /usr/src/debug/migraphx/AMDMIGraphX-rocm-7.2.2/src/include/migraphx/check_shapes.hpp:220: same_
dims: pointwise: Dimensions do not match
Dump: "/tmp/migraphx/fuse_pointwise_reduce57024748904648.mxr"
migraphx_program_compile: Error: /usr/src/debug/migraphx/AMDMIGraphX-rocm-7.2.2/src/include/migraphx/check_shapes.hpp:220: s
ame_dims: pointwise: Dimensions do not match
2026-05-10 03:19:23.781696378 [E:onnxruntime:, inference_session.cc:2625 operator()] Exception during initialization: Failed
to call function
Error: Error { code: RuntimeException, message: "Exception during initialization: Failed to call function", ptr: 0x55d8b4f48
e50 }
I was hoping you guys could give me some insight on why this isn't working. Thanks!
To reproduce
git clone https://github.com/jclsn/stemgen
git checkout feat/migraphx-support
export ORT_MIGRAPHX_MODEL_CACHE_PATH="$HOME/.cache/migraphx/models"
export ORT_MIGRAPHX_CACHE_PATH="$HOME/.cache/migraphx/"
cargo build --release --features migraphx && cargo build --release -p stemgen-cli --features migraphx
target/release/stemgen generate --device migraphx testdata/Oddchap\ -\ Sound\ 104.mp3 .
Urgency
Not urgent
Platform
Linux
OS Version
Arch Linux
ONNX Runtime Installation
Released Package
ONNX Runtime Version or Commit ID
1.24.4-6
ONNX Runtime API
C
Architecture
X64
Execution Provider
MIGraphX
Execution Provider Library Version
7.2.2-1
Describe the issue
I am currently trying to add support for AMD GPUs in an onnxruntime-based project. I am using the ort wrapper for Rust. For some reason model loading fails. There are three initial spikes on the CPU followed by small ones on the GPU. Then one core keeps being maxed out at almost 100%. Nothing happens for two hours then it fails.
It works well on the CPU and CUDA. I am pretty new to machine learning. Here is the call
https://github.com/jclsn/stemgen/blob/c41cb010fa8bbd458723623264305e66fb87569d/src/demucs.rs#L126
and the log
I was hoping you guys could give me some insight on why this isn't working. Thanks!
To reproduce
Urgency
Not urgent
Platform
Linux
OS Version
Arch Linux
ONNX Runtime Installation
Released Package
ONNX Runtime Version or Commit ID
1.24.4-6
ONNX Runtime API
C
Architecture
X64
Execution Provider
MIGraphX
Execution Provider Library Version
7.2.2-1