Modelpublic

nkkbr/whisper-large-v3-zatoichi-ja-EX-3-TRAIN_2_TO_10_EVAL_1_COSINE

sourceHugging Faceapache-2.0updated 10mo agoView on Hugging Face

0likes6downloads

Model Card

Whisper Large v3 - Japanese Zatoichi ASR

This model is a fine-tuned version of openai/whisper-large-v3 on the None dataset. It achieves the following results on the evaluation set:

—Loss: 0.5424
—Wer: 78.7573

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: 1e-05
—trainbatchsize: 32
—evalbatchsize: 8
—seed: 42
—optimizer: Use adamwtorchfused with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
—lrschedulertype: cosine
—lrschedulerwarmup_ratio: 0.1
—num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.3941	0.0217	1	1.4272	105.5416
1.5782	0.0435	2	1.0953	109.7397
1.0979	0.0652	3	1.0685	100.5877
1.1736	0.0870	4	1.0170	102.9387
1.1002	0.1087	5	0.9448	106.0453
1.1128	0.1304	6	0.9012	124.5172
1.0029	0.1522	7	0.8356	136.0202
0.9184	0.1739	8	0.8013	113.6860
0.9782	0.1957	9	0.7604	115.3652
0.8393	0.2174	10	0.7225	114.3577
0.7727	0.2391	11	0.6904	102.1830
0.6417	0.2609	12	0.6714	85.2225
0.6364	0.2826	13	0.6602	100.5877
0.7775	0.3043	14	0.6528	86.7338
0.7407	0.3261	15	0.6493	82.7876
0.6497	0.3478	16	0.6548	91.5197
0.6255	0.3696	17	0.6451	92.2754
0.6457	0.3913	18	0.6412	94.5424
0.6843	0.4130	19	0.6365	94.7943
0.59	0.4348	20	0.6313	93.7028
0.6106	0.4565	21	0.6183	91.2678
0.686	0.4783	22	0.6090	90.9320
0.6013	0.5	23	0.6058	81.1083
0.5634	0.5217	24	0.6025	77.5819
0.6321	0.5435	25	0.5963	77.3300
0.4556	0.5652	26	0.5948	87.1537
0.539	0.5870	27	0.5991	82.7036
0.5814	0.6087	28	0.6003	89.5886
0.7468	0.6304	29	0.5974	89.7565
0.5947	0.6522	30	0.5985	92.5273
0.6769	0.6739	31	0.5971	98.7406
0.612	0.6957	32	0.5930	95.0462
0.5619	0.7174	33	0.5853	96.0537
0.6309	0.7391	34	0.5771	94.9622
0.6389	0.7609	35	0.5694	94.3745
0.49	0.7826	36	0.5661	78.8413
0.5783	0.8043	37	0.5643	76.7422
0.551	0.8261	38	0.5614	75.3988
0.5752	0.8478	39	0.5564	76.0705
0.5175	0.8696	40	0.5532	75.8186
0.6194	0.8913	41	0.5517	80.1008
0.5258	0.9130	42	0.5515	79.7649
0.5516	0.9348	43	0.5501	79.6809
0.5352	0.9565	44	0.5496	77.1620
0.4998	0.9783	45	0.5479	77.8338
0.6088	1.0	46	0.5479	78.5055
0.3161	1.0217	47	0.5452	78.5055
0.3784	1.0435	48	0.5441	78.3375
0.3508	1.0652	49	0.5439	78.9253
0.3623	1.0870	50	0.5437	79.0932
0.3374	1.1087	51	0.5447	78.3375
0.4211	1.1304	52	0.5479	78.7573
0.3368	1.1522	53	0.5494	80.1008
0.4186	1.1739	54	0.5477	79.2611
0.3444	1.1957	55	0.5478	79.3451
0.3822	1.2174	56	0.5465	78.0856
0.3013	1.2391	57	0.5445	77.5819
0.2812	1.2609	58	0.5429	77.0781
0.3486	1.2826	59	0.5417	76.7422
0.3934	1.3043	60	0.5421	76.4903
0.3702	1.3261	61	0.5434	76.9941
0.386	1.3478	62	0.5460	77.4139
0.4185	1.3696	63	0.5495	80.1008
0.3125	1.3913	64	0.5527	80.4366
0.3822	1.4130	65	0.5543	81.0243
0.3812	1.4348	66	0.5543	81.1083
0.3552	1.4565	67	0.5528	81.1083
0.3072	1.4783	68	0.5509	81.0243
0.3348	1.5	69	0.5490	80.6885
0.2985	1.5217	70	0.5477	79.0932
0.2954	1.5435	71	0.5473	78.5894
0.292	1.5652	72	0.5468	78.3375
0.3363	1.5870	73	0.5467	78.1696
0.3935	1.6087	74	0.5463	78.5055
0.2915	1.6304	75	0.5455	77.9177
0.4162	1.6522	76	0.5452	78.0856
0.3673	1.6739	77	0.5445	78.0017
0.375	1.6957	78	0.5443	78.3375
0.347	1.7174	79	0.5441	77.9177
0.4194	1.7391	80	0.5439	78.4215
0.3016	1.7609	81	0.5432	78.5894
0.3168	1.7826	82	0.5430	78.4215
0.3667	1.8043	83	0.5428	78.5055
0.3252	1.8261	84	0.5425	78.4215
0.3271	1.8478	85	0.5424	78.6734
0.3075	1.8696	86	0.5423	78.8413
0.315	1.8913	87	0.5424	79.0092
0.3645	1.9130	88	0.5420	78.5894
0.3633	1.9348	89	0.5421	78.7573
0.3146	1.9565	90	0.5421	78.5894
0.4173	1.9783	91	0.5421	78.8413
0.5484	2.0	92	0.5424	78.7573

Framework versions

—Transformers 4.57.3
—Pytorch 2.9.1+cu128
—Datasets 4.4.1
—Tokenizers 0.22.1