eval_results.json

{
    "epoch": 2.9690721649484537,
    "eval_audio_cosine_sim": 0.5888153910636902,
    "eval_loss": 3.0872552394866943,
    "eval_runtime": 496.7716,
    "eval_samples": 12,
    "eval_samples_per_second": 0.024,
    "eval_steps_per_second": 0.024,
    "eval_text_cosine_sim": 0.3267640173435211
}