Makkoen/whisper-large-v3-cit-do015-wd0-lr3e-06-steps1200-CA

CA

This model is a fine-tuned version of openai/whisper-large-v3 on the 3309 CA-2024-11-28 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4321
• Wer Ortho: 23.2856
• Wer: 16.0797

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: 3e-06
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 4
• 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: 200
• training_steps: 1200
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.785	0.5369	100	0.5298	29.2975	21.3801
0.546	1.0738	200	0.4648	26.4263	18.8952
0.4399	1.6107	300	0.4377	25.2369	18.2070
0.4172	2.1477	400	0.4270	24.0940	17.0450
0.3578	2.6846	500	0.4219	23.7409	17.0272
0.3183	3.2215	600	0.4265	23.7781	16.5803
0.2929	3.7584	700	0.4187	23.6480	16.8127
0.2635	4.2953	800	0.4248	23.8710	16.8037
0.2511	4.8322	900	0.4238	23.7502	16.7322
0.236	5.3691	1000	0.4302	23.7316	16.5088
0.2121	5.9060	1100	0.4312	23.5923	16.4015
0.2089	6.4430	1200	0.4321	23.2856	16.0797

Framework versions

• Transformers 4.45.1
• Pytorch 1.13.1+cu117
• Datasets 3.0.1
• Tokenizers 0.20.0