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

Loss: 3.5889
Wer: 92.3967

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: 0.0002
train_batch_size: 4
eval_batch_size: 16
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 16
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: 500
training_steps: 10000
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.384	0.1797	250	0.9966	96.1662
0.434	0.3595	500	1.4886	98.5007
0.4014	0.5392	750	1.4760	97.7940
0.3318	0.7189	1000	1.5314	97.7511
0.3014	0.8986	1250	1.5504	97.8368
0.2213	1.0784	1500	1.6095	97.3656
0.2212	1.2581	1750	1.6825	96.1662
0.2323	1.4378	2000	1.5175	97.6012
0.2049	1.6175	2250	2.0035	97.7940
0.1834	1.7973	2500	1.6968	96.4232
0.2012	1.9770	2750	1.7613	97.3013
0.1426	2.1567	3000	1.5106	95.9734
0.1344	2.3364	3250	1.7199	95.5665
0.1512	2.5162	3500	1.9328	94.8169
0.1346	2.6959	3750	1.7806	96.0805
0.1211	2.8756	4000	2.0429	95.6736
0.0824	3.0554	4250	2.0699	95.3309
0.0936	3.2351	4500	2.0379	96.1876
0.0946	3.4148	4750	2.1346	95.9092
0.0904	3.5945	5000	2.1195	95.0311
0.0937	3.7743	5250	1.7738	95.1810
0.0836	3.9540	5500	2.0081	95.1167
0.0525	4.1337	5750	2.3687	94.9240
0.0562	4.3134	6000	2.2252	95.1381
0.0506	4.4932	6250	2.5513	95.5022
0.0592	4.6729	6500	2.5357	95.6736
0.0521	4.8526	6750	2.4758	95.8235
0.0276	5.0324	7000	2.8255	94.9454
0.0278	5.2121	7250	2.6255	94.7740
0.0311	5.3918	7500	3.0046	94.4956
0.0269	5.5715	7750	2.8301	94.7312
0.0242	5.7513	8000	2.8859	94.2386
0.0255	5.9310	8250	2.5873	93.4676
0.0157	6.1107	8500	3.4027	93.6175
0.0092	6.2904	8750	3.5842	93.6389
0.0118	6.4702	9000	3.2694	93.9602
0.0086	6.6499	9250	3.3464	93.5318
0.01	6.8296	9500	3.4414	93.4461
0.0065	7.0093	9750	3.3491	92.6108
0.002	7.1891	10000	3.5889	92.3967

Framework versions

Transformers 4.49.0
Pytorch 2.6.0+cu124
Datasets 3.3.2
Tokenizers 0.21.0

sagarchapara/whisper-small-tel

Whisper Telugu - Fine-tuned

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

sagarchapara/
whisper-small-tel