jpwoodbu/gpt-oss-20b-int4-ov

This model was created on 2026-01-15 using optimum-cli installed with:

$ pip install "openvino>=2025.4.1" "optimum-intel[openvino]>=1.21.0" "openvino-genai>=2025.4.1" accelerate

The conversion command line was:

$ optimum-cli export openvino --model openai/gpt-oss-20b \
  --task text-generation-with-past \
  --weight-format int4 \
  ./gpt-oss-20b-int4-ov

My experience running this on OpenVINO model server 2025.4.1 has had a few issues:

• The thinking text is delimited strangely and not recognized as thinking text by the two clients I've used: llm and chatbox.ai. This likely stems from the model using the Harmony Response Format.
• After just a few interactions, the output becomes repetative Chinese characters.

If anyone knows what might fix any of those issues, please let me know on the Community tab.