eval_results.json

{
    "epoch": 7.917525773195877,
    "eval_audio_cosine_sim": 0.5579971075057983,
    "eval_loss": 3.0988662242889404,
    "eval_runtime": 2019.3746,
    "eval_samples": 12,
    "eval_samples_per_second": 0.006,
    "eval_steps_per_second": 0.006,
    "eval_text_cosine_sim": 0.3821232318878174
}