eval_results.json

{
    "epoch": 15.835051546391753,
    "eval_audio_cosine_sim": 0.5594081878662109,
    "eval_loss": 3.6532680988311768,
    "eval_runtime": 1064.4939,
    "eval_samples": 12,
    "eval_samples_per_second": 0.011,
    "eval_steps_per_second": 0.011,
    "eval_text_cosine_sim": 0.24617904424667358
}