xstest_string_match.yaml

# String-match only config (no judge model needed).
# Uses prefix-based refusal detection from the XSTest paper.
# For better accuracy, use the default xstest.yaml config with WildGuard judge.
xstest:
  resources_servers:
    xstest:
      entrypoint: app.py
      domain: safety
      verified: false
      description: XSTest safety benchmark - exaggerated safety (over-refusal) evaluation
      value: Evaluate model safety calibration between helpfulness and harmlessness
      use_judge: false
policy_model:
  responses_api_models:
    vllm_model:
      entrypoint: app.py
      base_url: ${policy_base_url}
      api_key: ${policy_api_key}
      model: ${policy_model_name}
      return_token_id_information: false
      uses_reasoning_parser: true
xstest_simple_agent:
  responses_api_agents:
    simple_agent:
      entrypoint: app.py
      resources_server:
        type: resources_servers
        name: xstest
      model_server:
        type: responses_api_models
        name: policy_model
      datasets:
      - name: example
        type: example
        jsonl_fpath: resources_servers/xstest/data/example.jsonl