eval_results.json

{
    "epoch": 2.6879999999999997,
    "eval_audio_cosine_sim": 0.632674515247345,
    "eval_loss": 3.0996882915496826,
    "eval_runtime": 2067.9427,
    "eval_samples": 12,
    "eval_samples_per_second": 0.006,
    "eval_steps_per_second": 0.006,
    "eval_text_cosine_sim": 0.4172757863998413
}