all_results.json

{
    "epoch": 3.0,
    "eval_audio_cosine_sim": 0.32668301463127136,
    "eval_loss": 5.101603984832764,
    "eval_runtime": 150.73,
    "eval_samples": 9,
    "eval_samples_per_second": 0.06,
    "eval_steps_per_second": 0.02,
    "eval_text_cosine_sim": 0.09613808989524841,
    "total_flos": 6235958582540448.0,
    "train_loss": 7.227042491989907,
    "train_runtime": 3530.6424,
    "train_samples": 1580,
    "train_samples_per_second": 1.343,
    "train_steps_per_second": 0.084
}