whisper-nm-nomimose-ag

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: 0.2483
• Wer: 21.2742

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: 40
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	1.7009	100	0.4217	32.9920
0.9886	3.3932	200	0.3015	275.9954
0.9886	5.0855	300	0.2714	67.1217
0.2263	6.7863	400	0.2278	34.1297
0.2263	8.4786	500	0.2648	314.3345
0.1309	10.1709	600	0.2952	132.0819
0.1309	11.8718	700	0.2093	131.0580
0.0924	13.5641	800	0.3086	161.3197
0.0924	15.2564	900	0.2621	30.9443
0.0739	16.9573	1000	0.2176	30.0341
0.0739	18.6496	1100	0.2371	33.9022
0.0433	20.3419	1200	0.2281	33.9022
0.0433	22.0342	1300	0.2411	33.1058
0.0249	23.7350	1400	0.2423	28.6689
0.0249	25.4274	1500	0.2758	32.5370
0.0135	27.1197	1600	0.2588	27.8726
0.0135	28.8205	1700	0.2683	28.4414
0.0058	30.5128	1800	0.2603	23.4357
0.0058	32.2051	1900	0.2485	20.9329
0.0003	33.9060	2000	0.2483	21.2742
0.0003	35.5983	2100	0.2482	21.2742
0.0	37.2906	2200	0.2483	21.2742
0.0	38.9915	2300	0.2483	21.2742

Framework versions

• Transformers 4.47.0.dev0
• Pytorch 2.4.0
• Datasets 3.0.1
• Tokenizers 0.20.0