microsoft / onnxruntime-genai

Generative AI extensions for onnxruntime
MIT License
497 stars 124 forks source link

Issues running on Ryzen #728

Open jarroddavis68 opened 3 months ago

jarroddavis68 commented 3 months ago

Using phi3 vision model:

A client sees this warning on their machine for text generation, still works:

Unable to create a device from version 1.614.0 of the DirectX 12 Agility SDK.
You can still use this library, but some scenarios may not work.

When using vision, they get this additional error:

024-07-26 21:15:53.1830129 [E:onnxruntime:onnxruntime-genai, sequential_executor.cc:516
onnxruntime::ExecuteKernel] Non-zero status code returned while running MemcpyToHost node.
Name:'Memcpy_token_5' Status Message:
Error: Non-zero status code returned while running MemcpyToHost node. Name:'Memcpy_token_5'

Machine specs:

Ryzen 7 4700u, 16gb ram, igpu Vega 7, ssd, Windows 11
anujj commented 2 months ago

getting the same error with llama 3.1 model

skyline75489 commented 2 months ago

@anujj Are you also using DML? Could you please list your GPU spec? Thanks

natke commented 2 months ago

@anujj Please list your operating system version, GPU specs and which version of onnxruntime genai are you running?