josemhernandezbiometric/whisper-medium-finetuned-int8

Whisper Small Es - Sanchit Gandhi

This model is a fine-tuned version of openai/whisper-small on the Common Voice 11.0 dataset. It achieves the following results on the evaluation set:

• Loss: 2.5129
• Wer: 56.4413

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: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 256
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• training_steps: 750
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
6.3605	2.3	50	6.2660	55.7247
5.3113	4.6	100	5.1187	56.4590
4.2749	6.9	150	4.2391	55.6185
3.5266	9.2	200	3.4143	53.6719
3.0671	11.49	250	3.1045	49.2037
2.8716	13.79	300	2.9260	50.7786
2.7263	16.09	350	2.7987	53.5746
2.6467	18.39	400	2.7079	55.0787
2.5624	20.69	450	2.6443	55.6008
2.5087	22.99	500	2.5989	57.3881
2.4922	25.29	550	2.5660	55.9370
2.4274	27.59	600	2.5421	56.4325
2.4337	29.89	650	2.5257	57.4058
2.3991	32.18	700	2.5165	57.0165
2.4211	34.48	750	2.5129	56.4413

Framework versions

• PEFT 0.7.1
• Transformers 4.37.0.dev0
• Pytorch 2.1.0
• Datasets 2.16.2.dev0
• Tokenizers 0.15.0