UsmanAXAI/whisper-small-ft-client

Whisper small Ar - AxAI

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

• Loss: 1.5990
• Wer: 84.1146

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: 2
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.8044	6.37	200	1.2417	69.9219
0.036	12.75	400	1.1791	60.9375
0.0108	19.12	600	1.3128	80.2083
0.0035	25.5	800	1.3641	62.6953
0.0009	31.87	1000	1.4066	66.6016
0.0004	38.25	1200	1.4410	64.5833
0.0003	44.62	1400	1.4712	63.3464
0.0002	51.0	1600	1.4927	63.6068
0.0002	57.37	1800	1.5102	67.1875
0.0002	63.75	2000	1.5254	66.6016
0.0001	70.12	2200	1.5393	77.8646
0.0001	76.49	2400	1.5512	77.9297
0.0001	82.87	2600	1.5616	77.7344
0.0001	89.24	2800	1.5710	83.1380
0.0001	95.62	3000	1.5791	88.0859
0.0001	101.99	3200	1.5854	88.1510
0.0001	108.37	3400	1.5910	88.0859
0.0001	114.74	3600	1.5953	84.1146
0.0001	121.12	3800	1.5978	84.1797
0.0001	127.49	4000	1.5990	84.1146

Framework versions

• Transformers 4.37.2
• Pytorch 2.2.0+cu121
• Datasets 2.17.1
• Tokenizers 0.15.2