-
Notifications
You must be signed in to change notification settings - Fork 3.5k
Open
Labels
ep:ROCmquestions/issues related to ROCm execution providerquestions/issues related to ROCm execution provider
Description
Describe the issue
Memory access fault occurs when running inference on genderage.onnx
model with ROCMExecutionProvider on AMD Radeon RX 6800M.
To reproduce
Minimal reproduction script:
import numpy as np
import onnxruntime as ort
# Using insightface's genderage model
model_path = "genderage.onnx"
providers = ["ROCMExecutionProvider", "CPUExecutionProvider"]
session = ort.InferenceSession(model_path, providers=providers)
input_data = np.random.randn(1, 3, 96, 96).astype(np.float32)
input_name = session.get_inputs()[0].name
# This crashes with memory fault
outputs = session.run(None, {input_name: input_data})
Error:
Memory access fault by GPU node-1 (Agent handle: 0x55664a09cea0) on address 0x7f7ee8657000.
Reason: Page not present or supervisor privilege.
Urgency
No response
Platform
Linux
OS Version
Nobara Linux 42 (Fedora-based)
ONNX Runtime Installation
Released Package
ONNX Runtime Version or Commit ID
onnxruntime-rocm: 1.22.2.post1
ONNX Runtime API
Python
Architecture
X64
Execution Provider
Other / Unknown
Execution Provider Library Version
ROCm 6.3.1
Metadata
Metadata
Assignees
Labels
ep:ROCmquestions/issues related to ROCm execution providerquestions/issues related to ROCm execution provider