Skip to content

OpenVINO GenAI tests NPU support and Windows fixes#1660

Open
helena-intel wants to merge 2 commits intohuggingface:mainfrom
helena-intel:helena/test-ov-genai-npu
Open

OpenVINO GenAI tests NPU support and Windows fixes#1660
helena-intel wants to merge 2 commits intohuggingface:mainfrom
helena-intel:helena/test-ov-genai-npu

Conversation

@helena-intel
Copy link
Copy Markdown
Collaborator

Update OpenVINO GenAI tests

  • Fix issues and access violations caused by TemporaryDirectory on Windows
  • Add initial support for NPU
    • Speech2Text (whisper), selected LLMs and selected VLMs are supported for now. More models will be added later.
  • For LLMs, compare tokens instead of detokenized text. This fixes issues on GPU
  • On GPU, there are a few known failures, mentioned at the top of the file. We are looking into this.
  • Use chat template in VLM tests, which is also done in preprocess_inputs in optimum-intel (if models added in the future will not have this we can make this an option but for now all VLM tests pass with this)

The solution for the temporary directory looks convoluted but this was trickier than expected when we also want to delete the directory if the test fails.

I tested GPU and NPU on LNL 258V with Linux and Windows.

- Fix TemporaryDirectory issues on Windows
- Compare model output tokens instead of tokenized outputs for LLMs
- Initial NPU support
- Use chat template for VLM test
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant