whisper-base-1m.hr

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

• Loss: 0.2480
• Wer: 17.3979

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: 6.25e-06
• train_batch_size: 64
• eval_batch_size: 32
• 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: 800
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2907	0.0650	1000	0.3627	25.6512
0.2591	0.1300	2000	0.3494	26.1731
0.2394	0.1950	3000	0.3402	24.5520
0.2234	0.2600	4000	0.3332	24.0163
0.2241	0.3250	5000	0.3254	23.9978
0.2173	0.3900	6000	0.3178	23.2126
0.2108	0.4550	7000	0.3121	22.7970
0.2129	0.5200	8000	0.3069	22.2520
0.1993	0.5849	9000	0.3011	22.4044
0.2009	0.6499	10000	0.2968	21.8917
0.2	0.7149	11000	0.2930	21.6747
0.1981	0.7799	12000	0.2903	21.4391
0.1915	0.8449	13000	0.2879	20.4184
0.1902	0.9099	14000	0.2836	21.3052
0.1881	0.9749	15000	0.2803	20.9449
0.1706	1.0399	16000	0.2805	20.8064
0.1746	1.1049	17000	0.2787	20.7833
0.1716	1.1699	18000	0.2782	19.5825
0.172	1.2349	19000	0.2757	20.2476
0.1666	1.2999	20000	0.2740	20.2845
0.1574	1.3649	21000	0.2737	19.4763
0.1668	1.4299	22000	0.2705	19.5317
0.1697	1.4949	23000	0.2696	19.7210
0.1593	1.5599	24000	0.2665	19.3977
0.1665	1.6249	25000	0.2659	19.0052
0.1592	1.6898	26000	0.2653	18.3863
0.1629	1.7548	27000	0.2608	18.9775
0.1588	1.8198	28000	0.2612	18.9544
0.1559	1.8848	29000	0.2615	19.6379
0.1552	1.9498	30000	0.2590	18.3909
0.1452	2.0148	31000	0.2593	18.1969
0.1485	2.0798	32000	0.2590	18.1554
0.1464	2.1448	33000	0.2590	18.5433
0.1446	2.2098	34000	0.2583	17.9060
0.1361	2.2748	35000	0.2573	18.6265
0.1434	2.3398	36000	0.2585	18.1969
0.1426	2.4048	37000	0.2557	17.6935
0.1497	2.4698	38000	0.2552	17.9845
0.14	2.5348	39000	0.2558	18.3586
0.1428	2.5998	40000	0.2549	17.2640
0.147	2.6648	41000	0.2539	18.0260
0.1411	2.7298	42000	0.2531	17.4626
0.1441	2.7947	43000	0.2528	18.8989
0.1521	2.8597	44000	0.2518	17.6889
0.139	2.9247	45000	0.2515	18.4925
0.1402	2.9897	46000	0.2507	17.3102
0.1337	3.0547	47000	0.2522	17.7120
0.131	3.1197	48000	0.2526	17.7674
0.1282	3.1847	49000	0.2530	17.7028
0.1289	3.2497	50000	0.2520	17.7258
0.1296	3.3147	51000	0.2514	18.3863
0.132	3.3797	52000	0.2512	17.9845
0.1354	3.4447	53000	0.2512	17.2963
0.13	3.5097	54000	0.2505	17.5734
0.1311	3.5747	55000	0.2507	17.7443
0.133	3.6397	56000	0.2491	16.7652
0.1317	3.7047	57000	0.2488	17.0931
0.1335	3.7697	58000	0.2489	18.2847
0.1334	3.8347	59000	0.2489	17.7305
0.1287	3.8996	60000	0.2480	17.8413
0.1259	3.9646	61000	0.2481	17.4210
0.1261	4.0296	62000	0.2495	17.3887
0.1247	4.0946	63000	0.2498	17.4995
0.1207	4.1596	64000	0.2495	17.3564
0.1219	4.2246	65000	0.2494	17.3102
0.1219	4.2896	66000	0.2493	17.1670
0.1258	4.3546	67000	0.2491	17.3148
0.1266	4.4196	68000	0.2488	17.1993
0.1267	4.4846	69000	0.2489	17.2132
0.1267	4.5496	70000	0.2483	17.3471
0.1245	4.6146	71000	0.2482	17.5088
0.126	4.6796	72000	0.2482	17.2317
0.122	4.7446	73000	0.2482	17.1301
0.1288	4.8096	74000	0.2483	17.3564
0.1254	4.8746	75000	0.2481	17.2409
0.1313	4.9396	76000	0.2480	17.3979

Framework versions

• Transformers 4.47.0
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.21.0