__g__d____model

This model is a fine-tuned version of openai/whisper-medium on the whsNect/g__d_ dataset. It achieves the following results on the evaluation set:

• Loss: 0.0502
• Wer: 8.4602

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: 5e-06
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• optimizer: Use adamw_bnb_8bit 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: 500
• training_steps: 15000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0361	1.6722	500	0.0385	9.2003
0.0099	3.3445	1000	0.0313	5.2457
0.006	5.0167	1500	0.0335	6.3769
0.003	6.6890	2000	0.0348	4.8773
0.0021	8.3612	2500	0.0351	17.5822
0.0013	10.0334	3000	0.0369	5.0892
0.0016	11.7057	3500	0.0371	10.6837
0.0011	13.3779	4000	0.0367	5.8716
0.0014	15.0502	4500	0.0385	46.1350
0.0008	16.7224	5000	0.0408	10.2338
0.0006	18.3946	5500	0.0400	9.9077
0.0007	20.0669	6000	0.0410	11.2053
0.0003	21.7391	6500	0.0414	22.9192
0.0002	23.4114	7000	0.0415	17.6768
0.0009	25.0836	7500	0.0420	22.1074
0.0005	26.7559	8000	0.0440	14.8828
0.0005	28.4281	8500	0.0417	10.4065
0.0001	30.1003	9000	0.0441	20.4545
0.0001	31.7726	9500	0.0453	9.3176
0.0001	33.4448	10000	0.0460	11.3553
0.0001	35.1171	10500	0.0466	10.9999
0.0001	36.7893	11000	0.0471	11.0749
0.0001	38.4615	11500	0.0479	12.3887
0.0	40.1338	12000	0.0483	10.3413
0.0	41.8060	12500	0.0487	8.3363
0.0001	43.4783	13000	0.0491	8.6852
0.0	45.1505	13500	0.0495	7.7462
0.0	46.8227	14000	0.0499	8.1472
0.0	48.4950	14500	0.0501	7.9516
0.0	50.1672	15000	0.0502	8.4602

Framework versions

• Transformers 4.49.0
• Pytorch 2.2.2+cu121
• Datasets 3.4.1
• Tokenizers 0.21.1