eval_results.json

{
    "epoch": 15.835051546391753,
    "eval_audio_cosine_sim": 0.5780665278434753,
    "eval_loss": 3.6437976360321045,
    "eval_runtime": 1037.2655,
    "eval_samples": 12,
    "eval_samples_per_second": 0.012,
    "eval_steps_per_second": 0.012,
    "eval_text_cosine_sim": 0.246041402220726
}