whisper-child-75

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

• Loss: 0.3811
• Wer: 19.0794

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: 1e-05
• train_batch_size: 48
• eval_batch_size: 32
• seed: 42
• optimizer: Use OptimizerNames.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_steps: 70
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.0406	0.1078	25	1.2208	38.0682
1.0219	0.2155	50	1.1918	37.6556
0.9715	0.3233	75	1.1313	36.7397
0.9105	0.4310	100	1.0425	35.4715
0.8408	0.5388	125	0.9366	34.8576
0.7305	0.6466	150	0.8256	32.5058
0.6843	0.7543	175	0.7052	31.1974
0.5954	0.8621	200	0.6055	28.8187
0.5398	0.9698	225	0.5391	25.5074
0.5099	1.0776	250	0.4902	23.0952
0.4845	1.1853	275	0.4555	22.3236
0.4858	1.2931	300	0.4344	21.1897
0.4741	1.4009	325	0.4224	20.9213
0.4589	1.5086	350	0.4143	20.1161
0.4294	1.6164	375	0.4081	20.6428
0.426	1.7241	400	0.4027	21.2467
0.406	1.8319	425	0.3984	20.2469
0.4443	1.9397	450	0.3948	19.9416
0.4351	2.0474	475	0.3920	20.7804
0.4394	2.1552	500	0.3897	21.1393
0.4167	2.2629	525	0.3874	19.6196
0.3827	2.3707	550	0.3855	19.3981
0.4164	2.4784	575	0.3842	19.1767
0.4046	2.5862	600	0.3830	18.9855
0.4196	2.6940	625	0.3821	18.9184
0.4008	2.8017	650	0.3814	19.0861
0.3902	2.9095	675	0.3811	19.0794

Framework versions

• PEFT 0.16.0
• Transformers 4.52.0
• Pytorch 2.7.1+cu126
• Datasets 3.6.0
• Tokenizers 0.21.2