Modelpublic

nkkbr/whisper-large-v3-zatoichi-ja-EX-4-TRAIN_2_TO_10_EVAL_1_COSINE

sourceHugging Faceapache-2.0updated 10mo agoView on Hugging Face

0likes5downloads

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: 0.5258
—Wer: 78.8194

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
1.3938	0.0263	1	1.3281	98.2639
1.4874	0.0526	2	1.0483	96.2674
1.0229	0.0789	3	1.0172	102.5174
1.0766	0.1053	4	0.9644	98.8715
1.1038	0.1316	5	0.8872	100.2604
0.9138	0.1579	6	0.8196	106.1632
0.937	0.1842	7	0.7850	109.2014
0.9006	0.2105	8	0.7481	113.2812
0.9447	0.2368	9	0.7279	113.1076
0.6762	0.2632	10	0.6883	111.3715
0.8225	0.2895	11	0.6641	106.8576
0.702	0.3158	12	0.6420	80.5556
0.6451	0.3421	13	0.6337	81.8576
0.7005	0.3684	14	0.6276	81.1632
0.5502	0.3947	15	0.6232	85.8507
0.622	0.4211	16	0.6217	86.5451
0.6595	0.4474	17	0.6178	81.4236
0.6755	0.4737	18	0.6093	80.2951
0.6307	0.5	19	0.5993	84.7222
0.5655	0.5263	20	0.5935	92.1875
0.6415	0.5526	21	0.5917	91.3194
0.6899	0.5789	22	0.5910	85.3299
0.565	0.6053	23	0.5858	90.2778
0.5625	0.6316	24	0.5816	91.3194
0.6392	0.6579	25	0.5798	85.3299
0.4879	0.6842	26	0.5732	83.3333
0.5908	0.7105	27	0.5663	85.9375
0.5397	0.7368	28	0.5614	85.8507
0.5265	0.7632	29	0.5576	86.6319
0.5436	0.7895	30	0.5550	80.7292
0.5772	0.8158	31	0.5526	81.6840
0.5597	0.8421	32	0.5515	83.3333
0.5789	0.8684	33	0.5486	84.9826
0.5546	0.8947	34	0.5448	91.0590
0.5935	0.9211	35	0.5402	85.8507
0.5417	0.9474	36	0.5365	85.1562
0.5152	0.9737	37	0.5344	82.8125
0.5517	1.0	38	0.5331	82.7257
0.3677	1.0263	39	0.5311	76.9965
0.4679	1.0526	40	0.5316	78.2118
0.4012	1.0789	41	0.5322	83.3333
0.3178	1.1053	42	0.5324	86.2847
0.3857	1.1316	43	0.5338	86.1111
0.3564	1.1579	44	0.5362	86.7188
0.4102	1.1842	45	0.5391	82.8125
0.394	1.2105	46	0.5405	79.4271
0.3298	1.2368	47	0.5420	84.7222
0.3596	1.2632	48	0.5416	83.1597
0.3013	1.2895	49	0.5426	82.9861
0.381	1.3158	50	0.5423	82.8993
0.3631	1.3421	51	0.5412	83.5069
0.32	1.3684	52	0.5404	84.2014
0.5002	1.3947	53	0.5397	84.1146
0.3926	1.4211	54	0.5391	77.9514
0.3287	1.4474	55	0.5378	84.4618
0.3585	1.4737	56	0.5356	84.2882
0.3313	1.5	57	0.5333	77.7778
0.309	1.5263	58	0.5313	77.6042
0.3046	1.5526	59	0.5293	77.8646
0.3117	1.5789	60	0.5280	78.2118
0.3744	1.6053	61	0.5273	78.2118
0.3568	1.6316	62	0.5272	78.4722
0.3121	1.6579	63	0.5266	78.5590
0.3603	1.6842	64	0.5263	78.2118
0.3793	1.7105	65	0.5264	78.9931
0.2817	1.7368	66	0.5267	79.6007
0.3488	1.7632	67	0.5268	79.6875
0.2984	1.7895	68	0.5270	79.8611
0.3055	1.8158	69	0.5267	79.6875
0.3862	1.8421	70	0.5265	78.9062
0.3317	1.8684	71	0.5264	78.7326
0.2911	1.8947	72	0.5260	78.7326
0.2816	1.9211	73	0.5259	78.9062
0.3384	1.9474	74	0.5259	78.8194
0.3433	1.9737	75	0.5259	78.9062
0.406	2.0	76	0.5258	78.8194

Framework versions

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