gsm8k_aviary_resources_server:
  resources_servers:
    aviary:
      entrypoint: gsm8k_app.py
      domain: math
      verified: false
      description: GSM8k benchmark with calculator tool
      value: Test math and agentic capability
gsm8k_aviary_agent:
  responses_api_agents:
    aviary_agent:
      entrypoint: app.py
      resources_server:
        type: resources_servers
        name: gsm8k_aviary_resources_server
      model_server:
        type: responses_api_models
        name: policy_model
      max_steps: 10
      datasets:
      - name: train
        type: train
        jsonl_fpath: resources_servers/aviary/data/gsm8k_train.jsonl
        gitlab_identifier:
          dataset_name: PLACEHOLDER
          version: 0.0.1
          artifact_fpath: train.jsonl
        license: Apache 2.0
      - name: validation
        type: validation
        jsonl_fpath: resources_servers/aviary/data/gsm8k_validation.jsonl
        gitlab_identifier:
          dataset_name: PLACEHOLDER
          version: 0.0.1
          artifact_fpath: train.jsonl
        license: Apache 2.0
      - name: example
        type: example
        jsonl_fpath: resources_servers/aviary/data/gsm8k_example.jsonl