whisper-tiny-no-specific-topic-v5

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

• Loss: 0.8213
• Wer: 42.1182

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: 8
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine_with_restarts
• lr_scheduler_warmup_ratio: 0.1
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.3891	0.05	200	1.8372	86.0455
0.7174	0.1	400	0.8617	49.7
0.5472	0.15	600	0.7714	51.5727
0.5792	0.2	800	0.7486	45.6909
0.5513	0.25	1000	0.7440	45.7636
0.4622	0.3	1200	0.7561	44.9
0.3441	0.35	1400	0.7608	43.1273
0.3292	0.4	1600	0.7845	48.1182
0.294	0.45	1800	0.7794	43.3909
0.3153	0.5	2000	0.8136	45.6636
0.2735	0.55	2200	0.8016	45.7
0.2964	0.6	2400	0.7894	42.8636
0.2526	0.65	2600	0.8110	42.0455
0.3146	0.7	2800	0.8110	42.1364
0.2511	0.75	3000	0.8199	43.3636
0.2584	0.8	3200	0.8233	40.5909
0.2211	0.85	3400	0.8264	42.2727
0.266	0.9	3600	0.8193	42.1091
0.227	0.95	3800	0.8206	42.0182
0.2338	1.0	4000	0.8213	42.1182

Framework versions

• Transformers 4.49.0
• Pytorch 2.5.1+cu121
• Datasets 3.3.2
• Tokenizers 0.21.0