whisper-medium-tp

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

• Loss: 1.4684
• Wer Ortho: 33.3779
• Wer: 23.4989

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: 0.0001
• train_batch_size: 64
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 128
• 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: constant_with_warmup
• lr_scheduler_warmup_steps: 50
• training_steps: 300
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
1.8131	5.0	5	1.3364	33.8681	25.2700
1.2175	10.0	10	0.8872	36.4082	26.0475
0.6716	15.0	15	0.7097	56.5508	48.5529
0.2076	20.0	20	0.8878	64.8396	58.6177
0.0205	25.0	25	1.1002	36.4528	26.2635
0.0018	30.0	30	1.2256	36.0963	26.5227
0.0009	35.0	35	1.3132	50.0446	39.9136
0.0278	40.0	40	1.1617	29.1889	19.4816
0.0104	45.0	45	1.1375	30.7487	20.4320
0.0027	50.0	50	1.1603	30.5258	20.7343
0.0009	55.0	55	1.2324	40.8645	30.1512
0.0004	60.0	60	1.2813	33.3333	22.0302
0.0002	65.0	65	1.3132	32.1747	22.0734
0.0001	70.0	70	1.3377	34.7594	24.5356
0.0001	75.0	75	1.3560	28.9661	18.6609
0.0001	80.0	80	1.3700	29.0553	18.7473
0.0001	85.0	85	1.3796	29.9911	19.6544
0.0001	90.0	90	1.3861	29.7683	19.3521
0.0001	95.0	95	1.3903	29.7237	19.3521
0.0	100.0	100	1.3928	29.9020	19.6976
0.0	105.0	105	1.3944	29.9020	19.7408
0.0	110.0	110	1.3952	29.5900	19.4816
0.0	115.0	115	1.3956	29.5900	19.6112
0.0	120.0	120	1.3952	29.5900	19.6112
0.0	125.0	125	1.3948	29.7237	19.7840
0.0	130.0	130	1.3940	29.5900	19.6976
0.0	135.0	135	1.3927	29.6791	19.6976
0.0	140.0	140	1.3918	29.5900	19.4816
0.0	145.0	145	1.3906	29.8128	19.5248
0.0	150.0	150	1.3896	29.0998	18.9201
0.0	155.0	155	1.3883	28.9216	18.7473
0.0	160.0	160	1.3873	28.6542	18.8769
0.0	165.0	165	1.3866	28.6542	18.8769
0.0	170.0	170	1.3865	28.7433	18.9633
0.0	175.0	175	1.3865	28.7433	18.9633
0.0	180.0	180	1.3871	31.1052	21.1663
0.0	185.0	185	1.3884	31.1052	21.1663
0.0	190.0	190	1.3902	31.5954	21.4687
0.0	195.0	195	1.3926	31.5508	21.4255
0.0	200.0	200	1.3947	31.5062	21.4255
0.0	205.0	205	1.3976	31.5062	21.4255
0.0	210.0	210	1.4001	31.5508	21.5119
0.0	215.0	215	1.4032	31.7291	21.5551
0.0	220.0	220	1.4062	31.7736	21.5551
0.0	225.0	225	1.4091	31.7291	21.5551
0.0	230.0	230	1.4123	31.5954	21.5551
0.0	235.0	235	1.4158	31.5954	21.5551
0.0	240.0	240	1.4192	31.6399	21.6847
0.0	245.0	245	1.4230	32.3975	22.4622
0.0	250.0	250	1.4269	32.4421	22.3758
0.0	255.0	255	1.4308	32.2638	22.1598
0.0	260.0	260	1.4347	32.2638	22.1598
0.0	265.0	265	1.4385	31.6845	21.8575
0.0	270.0	270	1.4426	31.7291	21.9006
0.0	275.0	275	1.4466	32.3529	22.4622
0.0	280.0	280	1.4510	32.3529	22.4622
0.0	285.0	285	1.4551	32.0410	22.1598
0.0	290.0	290	1.4596	32.0410	22.1598
0.0	295.0	295	1.4640	33.2442	23.4125
0.0	300.0	300	1.4684	33.3779	23.4989

Framework versions

• Transformers 4.57.1
• Pytorch 2.9.0+cu128
• Datasets 4.2.0
• Tokenizers 0.22.1