eval_results.json

{
    "epoch": 1.0,
    "eval_audio_cosine_sim": 0.4679793119430542,
    "eval_loss": 3.913311243057251,
    "eval_runtime": 237.6952,
    "eval_samples": 9,
    "eval_samples_per_second": 0.038,
    "eval_steps_per_second": 0.013,
    "eval_text_cosine_sim": 0.23309490084648132
}