[Task]: Benchmark CLI Integration

[samet-akcay]

Description

Add CLI command for running benchmarks from configuration files. Integrates with jsonargparse for YAML-based composition.

Component

Library (VLA framework)

Acceptance Criteria

• uv run geti-action benchmark --config <path> command works
• Loads policy, gyms, and evaluation settings from YAML
• Supports checkpoint path or HuggingFace model ID
• Outputs results to specified directory
• Progress bar during evaluation
• Example YAML configs for LIBERO-10

Implementation Notes

# configs/benchmark/libero_eval.yaml
benchmark:
  task_suite: libero_10
  num_episodes: 20
  max_steps: 300

policy:
  checkpoint: ./checkpoints/act_libero.ckpt
  # OR: hub_id: open-edge-platform/act-libero-10

evaluation:
  video_dir: ./results/videos
  record_mode: failures
  output_dir: ./results

seed: 42

# CLI usage
uv run geti-action benchmark --config configs/benchmark/libero_eval.yaml

Files:

• library/src/getiaction/cli/benchmark.py (or extend existing CLI)
• library/configs/benchmark/libero_eval.yaml
• library/configs/benchmark/libero_quick.yaml
• library/tests/integration/test_benchmark_cli.py

Parent Epic

#161

Dependencies

No response