eval_results.json

{
    "epoch": 2.6879999999999997,
    "eval_audio_cosine_sim": 0.5482687950134277,
    "eval_loss": 3.092636823654175,
    "eval_runtime": 967.8097,
    "eval_samples": 12,
    "eval_samples_per_second": 0.012,
    "eval_steps_per_second": 0.012,
    "eval_text_cosine_sim": 0.4583999812602997
}