whisper_aii_clp

This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0177
• Wer: 16.1426

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: 0.0004
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use 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: 132
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	1.0	40	0.1406	27.2537
No log	2.0	80	0.3498	99.7904
1.105	3.0	120	0.2394	41.7191
1.105	4.0	160	0.2428	44.4444
0.1638	5.0	200	0.2523	52.8302
0.1638	6.0	240	0.1421	55.9748
0.1638	7.0	280	0.0707	33.7526
0.0797	8.0	320	0.0827	38.7841
0.0797	9.0	360	0.0417	14.8847
0.0379	10.0	400	0.0549	20.3354
0.0379	11.0	440	0.0750	28.9308
0.0379	12.0	480	0.0367	21.8029
0.0226	13.0	520	0.0358	18.2390
0.0226	14.0	560	0.0292	20.1258
0.0143	15.0	600	0.0192	22.0126
0.0143	16.0	640	0.0199	20.9644
0.0143	17.0	680	0.0198	20.5451
0.0072	18.0	720	0.0176	17.6101
0.0072	19.0	760	0.0175	16.7715
0.0072	19.5063	780	0.0177	16.1426

Framework versions

• Transformers 4.50.0.dev0
• Pytorch 2.5.1+cu121
• Datasets 3.3.1
• Tokenizers 0.21.0