eval_results.json

{
    "epoch": 0.896,
    "eval_audio_cosine_sim": 0.593883752822876,
    "eval_loss": 3.099468231201172,
    "eval_runtime": 2650.2034,
    "eval_samples": 12,
    "eval_samples_per_second": 0.005,
    "eval_steps_per_second": 0.005,
    "eval_text_cosine_sim": 0.41744545102119446
}