eval_results.json

{
    "epoch": 3.0,
    "eval_audio_cosine_sim": 0.32668301463127136,
    "eval_loss": 5.101603984832764,
    "eval_runtime": 150.73,
    "eval_samples": 9,
    "eval_samples_per_second": 0.06,
    "eval_steps_per_second": 0.02,
    "eval_text_cosine_sim": 0.09613808989524841
}