mcloud_oaieval.yaml

# Ultravox eval with vllm configuration
name: ultravox-oaieval
image: mosaicml/composer:latest
compute:
  gpus: 8
  cluster: r15z1p1
integrations:
  - integration_type: git_repo
    git_repo: fixie-ai/ultravox
    git_branch: $UV_BRANCH
    pip_install: poetry==1.7.1
scheduling:
  max_duration: 2  # 2 hours max for jobs to avoid hanging jobs
command: >-
  cd ultravox &&
  poetry install --no-dev &&
  poetry run pip install vllm &&
  HF_TOKEN=$HF_WRITE_TOKEN poetry run python -m ultravox.inference.run_vllm_inference $EVAL_ARGS
env_variables:
  MLFLOW_TRACKING_URI: databricks
  UV_BRANCH: main
  EVAL_ARGS: --evalset audio-core --model fixie-ai/ultravox-v0_5-llama-3_1-8b