distil-large-v2_ADReSSo

This model is a fine-tuned version of distil-whisper/distil-large-v2 on the audiofolder dataset. It achieves the following results on the evaluation set:

• Loss: 1.7772
• Accuracy: 0.7606
• F1: 0.7733

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 3e-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1
0.7	1.0	16	0.6865	0.5714	0.25
0.6796	2.0	32	0.6753	0.5	0.0
0.5951	3.0	48	0.5741	0.6905	0.6667
0.3828	4.0	64	1.2909	0.5476	0.24
0.1822	5.0	80	1.0336	0.7857	0.7907
0.2948	6.0	96	0.9771	0.7619	0.7222
0.064	7.0	112	0.9788	0.8333	0.8293
0.0007	8.0	128	1.0079	0.8333	0.8444
0.0003	9.0	144	1.0393	0.8333	0.8444
0.0002	10.0	160	1.0977	0.8571	0.8696
0.0001	11.0	176	1.1214	0.8571	0.8696
0.0001	12.0	192	1.1597	0.8571	0.8696
0.0001	13.0	208	1.1867	0.8571	0.8696
0.0001	14.0	224	1.2124	0.8571	0.8696
0.0001	15.0	240	1.2319	0.8571	0.8696
0.0001	16.0	256	1.2440	0.8571	0.8696
0.0001	17.0	272	1.2629	0.8571	0.8696
0.0001	18.0	288	1.2777	0.8571	0.8696
0.0001	19.0	304	1.2876	0.8571	0.8696
0.0	20.0	320	1.3026	0.8571	0.8696
0.0	21.0	336	1.3156	0.8571	0.8696
0.0	22.0	352	1.3261	0.8571	0.8696
0.0	23.0	368	1.3362	0.8571	0.8696
0.0	24.0	384	1.3445	0.8571	0.8696
0.0	25.0	400	1.3532	0.8571	0.8696
0.0	26.0	416	1.3604	0.8571	0.8696
0.0	27.0	432	1.3673	0.8571	0.8696
0.0	28.0	448	1.3746	0.8571	0.8696
0.0	29.0	464	1.3809	0.8571	0.8696
0.0	30.0	480	1.3878	0.8571	0.8696

Framework versions

• Transformers 4.51.3
• Pytorch 2.5.0+cu118
• Datasets 2.14.6
• Tokenizers 0.21.1