Closed Jose17-ml closed 3 days ago
Hi Experts,
Need your inputs.
Hi,
Any inputs?
Hi Josh, can you provide the information of you GPU device by sending a screenshot of from task manager performace tab. This model is not in our original supported model list. It's much likely the layernorm op used in this op is slightly different from DML-EP supports. It would be helpful if you could provide us more details of the layernorm op usage in this model.
Closing issue due to low activity. Feel free to reopen it with more information about the op.
Describe the feature request
Hi Experts,
I just started working AI/ML stuff recently. Currently trying to run Hugging Face - Optimum model on GPU using DML-EP
Platform: Windows 11
Model: https://huggingface.co/optimum/m2m100_418M
Changes:
import onnxruntime
session_opt = onnxruntime.SessionOptions() session_opt.log_severity_level = 0
provider = "CPUExecutionProvider"
provider = "DmlExecutionProvider" NUM_ITERATIONS = 1
model_name = "optimum/m2m100_418M"
hi_text = "जीवन एक चॉकलेट बॉक्स की तरह है।" chinese_text = "生活就像一盒巧克力。"
model = ORTModelForSeq2SeqLM.from_pretrained(model_name, provider=provider, session_options=session_opt)
When I use "DmlExecutionProvider", I see below error
2024-06-12 14:35:21.2694023 [E:onnxruntime:, sequential_executor.cc:516 onnxruntime::ExecuteKernel] Non-zero status code returned while running LayerNormalization node. Name:'/model/decoder/layer_norm/Mul/LayerNormFusion/' Status Message: C:\a_work\1\s\onnxruntime\core\providers\dml\DmlExecutionProvider\src\MLOperatorAuthorImpl.cpp(2468)\onnxruntime_pybind11_state.pyd!00007FFA9B5A09BF: (caller: 00007FFA9B5A2174) Exception(3) tid(1ff4) 887A0005 The GPU device instance has been suspended. Use GetDeviceRemovedReason to determine the appropriate action.
But where as with "CPUExecutionProvider", I don't see any issue and able to run the model successfully.
So, I need your help to resolve this issue and run with DML-EP.
Thanks
Describe scenario use case
Trying to huggingface-Optimum model with DML-EP