eval_results.json

{
    "epoch": 7.917525773195877,
    "eval_audio_cosine_sim": 0.5995970368385315,
    "eval_loss": 3.09893536567688,
    "eval_runtime": 1924.8031,
    "eval_samples": 12,
    "eval_samples_per_second": 0.006,
    "eval_steps_per_second": 0.006,
    "eval_text_cosine_sim": 0.39815211296081543
}