eval_results.json

{
    "epoch": 2.6879999999999997,
    "eval_audio_cosine_sim": 0.5799151062965393,
    "eval_loss": 3.0975494384765625,
    "eval_runtime": 1080.4689,
    "eval_samples": 12,
    "eval_samples_per_second": 0.011,
    "eval_steps_per_second": 0.011,
    "eval_text_cosine_sim": 0.49306759238243103
}