exp_008_base_cv24_vanilla

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

• Loss: 0.3305
• Wer: 27.1291
• Wer Ortho: 30.1229
• Cer: 8.7897

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: 5e-05
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 2000
• training_steps: 20000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Wer Ortho	Cer
0.5773	0.3445	1000	0.5443	48.3328	51.3254	15.6376
0.4567	0.6889	2000	0.4493	39.7212	42.6617	13.2815
0.3257	1.0334	3000	0.3916	34.2850	37.2906	11.0731
0.3101	1.3779	4000	0.3619	32.8548	35.8135	11.1400
0.2987	1.7224	5000	0.3424	31.0170	34.0501	10.4724
0.2019	2.0668	6000	0.3367	29.8753	32.8314	10.0224
0.2100	2.4113	7000	0.3303	28.9296	32.0212	9.7600
0.2068	2.7558	8000	0.3193	27.9595	31.0167	9.0057
0.1307	3.1002	9000	0.3343	27.9822	31.0302	9.1302
0.1381	3.4447	10000	0.3335	27.8125	30.8133	8.9957
0.1353	3.7892	11000	0.3305	27.1291	30.1229	8.7897
0.0784	4.1337	12000	0.3474	27.5965	30.7395	9.0974
0.0785	4.4781	13000	0.3553	27.3914	30.3821	8.9799
0.0761	4.8226	14000	0.3551	27.4994	30.4820	8.9953

Framework versions

• Transformers 5.2.0
• Pytorch 2.10.0+cu128
• Datasets 3.6.0
• Tokenizers 0.22.2