whisper-medium-sorani-v1

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: 0.2531
• Wer: 18.9917

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-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Use OptimizerNames.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: 1250
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2868	0.3365	500	0.3089	43.5663
0.2449	0.6729	1000	0.2722	41.1446
0.2169	1.0094	1500	0.2523	37.0698
0.1657	1.3459	2000	0.2198	33.5799
0.1622	1.6824	2500	0.2001	30.5878
0.0903	2.0188	3000	0.1891	28.5673
0.0997	2.3553	3500	0.1959	29.2150
0.1011	2.6918	4000	0.1738	27.6682
0.0605	3.0283	4500	0.1892	27.4459
0.0538	3.3647	5000	0.1953	27.0495
0.0662	3.7012	5500	0.1816	25.2417
0.0337	4.0377	6000	0.1968	25.4060
0.0372	4.3742	6500	0.1978	24.5698
0.0335	4.7106	7000	0.1993	23.8012
0.0225	5.0471	7500	0.2147	24.2556
0.0305	5.3836	8000	0.2007	23.8592
0.0279	5.7201	8500	0.2105	24.2846
0.0156	6.0565	9000	0.2077	22.9988
0.0173	6.3930	9500	0.2177	23.0278
0.0167	6.7295	10000	0.2148	22.7523
0.0118	7.0659	10500	0.2232	22.7523
0.0132	7.4024	11000	0.2185	23.2502
0.0171	7.7389	11500	0.2167	23.2115
0.0096	8.0754	12000	0.2233	22.6363
0.0106	8.4118	12500	0.2167	21.8581
0.0116	8.7483	13000	0.2227	22.4188
0.0074	9.0848	13500	0.2265	21.6067
0.0085	9.4213	14000	0.2305	22.0998
0.0107	9.7577	14500	0.2409	21.9499
0.0065	10.0942	15000	0.2258	21.1959
0.0058	10.4307	15500	0.2295	21.5922
0.0044	10.7672	16000	0.2343	21.5052
0.0041	11.1036	16500	0.2345	21.3312
0.0055	11.4401	17000	0.2276	21.3844
0.0035	11.7766	17500	0.2366	20.9735
0.0026	12.1131	18000	0.2387	20.4853
0.0036	12.4495	18500	0.2277	20.6255
0.0018	12.7860	19000	0.2396	20.5191
0.0025	13.1225	19500	0.2292	20.3258
0.0017	13.4590	20000	0.2385	20.3113
0.0017	13.7954	20500	0.2388	20.2533
0.0009	14.1319	21000	0.2399	20.0454
0.0017	14.4684	21500	0.2424	19.8231
0.0016	14.8048	22000	0.2437	20.1373
0.0005	15.1413	22500	0.2417	19.9923
0.0019	15.4778	23000	0.2399	19.3010
0.0006	15.8143	23500	0.2449	19.1899
0.0003	16.1507	24000	0.2518	19.1850
0.0006	16.4872	24500	0.2555	19.4026
0.0009	16.8237	25000	0.2468	19.3010
0.0011	17.1602	25500	0.2461	19.2769
0.0004	17.4966	26000	0.2418	19.2624
0.0001	17.8331	26500	0.2525	19.1125
0.0001	18.1696	27000	0.2509	19.0594
0.0	18.5061	27500	0.2520	19.0690
0.0001	18.8425	28000	0.2516	19.0497
0.0	19.1790	28500	0.2521	19.0449
0.0	19.5155	29000	0.2526	18.9869
0.0	19.8520	29500	0.2531	18.9917

Framework versions

• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.2