Modelpublic

dianavdavidson/indicwav2vec-hindi-vaani-62143-normalized-2e-5-epochs-100-FT

sourceHugging Faceapache-2.0updated 28d agoView on Hugging Face

0likes16downloads

Model Card

indicwav2vec-hindi-vaani-62143-normalized-2e-5-epochs-100-FT

This model is a fine-tuned version of ai4bharat/indicwav2vec-hindi on an unknown dataset. It achieves the following results on the evaluation set:

—Loss: 0.3157
—Global Wer: 15.1236

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: 2e-05
—trainbatchsize: 16
—evalbatchsize: 16
—seed: 42
—gradientaccumulationsteps: 2
—totaltrainbatch_size: 32
—optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
—lrschedulertype: constantwithwarmup
—lrschedulerwarmup_steps: 500
—num_epochs: 100
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Global Wer
13.0362	1.0	540	1.8982	99.9962
2.3876	2.0	1080	0.6386	42.7360
1.3937	3.0	1620	0.5195	36.5124
1.1913	4.0	2160	0.4636	33.1046
1.0724	5.0	2700	0.4288	30.8806
0.9963	6.0	3240	0.4059	28.8299
0.9350	7.0	3780	0.3852	27.8875
0.9003	8.0	4320	0.3699	26.5455
0.8436	9.0	4860	0.3613	25.9612
0.8155	10.0	5400	0.3486	25.3619
0.7885	11.0	5940	0.3414	24.7738
0.7624	12.0	6480	0.3351	24.1933
0.7235	13.0	7020	0.3261	23.6580
0.7012	14.0	7560	0.3240	23.0737
0.6848	15.0	8100	0.3153	22.9003
0.6632	16.0	8640	0.3123	22.3990
0.6467	17.0	9180	0.3080	22.0371
0.6434	18.0	9720	0.3043	21.9315
0.6197	19.0	10260	0.3024	21.3359
0.6038	20.0	10800	0.2994	21.2153
0.6006	21.0	11340	0.2932	21.2681
0.5743	22.0	11880	0.2935	20.7027
0.5689	23.0	12420	0.2903	20.5594
0.5628	24.0	12960	0.2913	20.4576
0.5375	25.0	13500	0.2863	20.1561
0.5291	26.0	14040	0.2836	19.9638
0.5214	27.0	14580	0.2785	19.9450
0.5083	28.0	15120	0.2810	19.8394
0.4983	29.0	15660	0.2775	19.5002
0.4987	30.0	16200	0.2748	19.5680
0.4842	31.0	16740	0.2750	19.2438
0.4776	32.0	17280	0.2746	18.9686
0.4697	33.0	17820	0.2726	19.0855
0.4615	34.0	18360	0.2741	18.7802
0.4507	35.0	18900	0.2720	18.7161
0.4467	36.0	19440	0.2764	18.3881
0.4399	37.0	19980	0.2699	18.3429
0.4236	38.0	20520	0.2679	18.2713
0.4187	39.0	21060	0.2740	18.0489
0.4117	40.0	21600	0.2717	17.9697
0.4045	41.0	22140	0.2724	18.0489
0.3961	42.0	22680	0.2736	17.7699
0.3895	43.0	23220	0.2733	17.7284
0.3894	44.0	23760	0.2708	17.4269
0.3796	45.0	24300	0.2725	17.3364
0.3701	46.0	24840	0.2772	17.1291
0.3660	47.0	25380	0.2767	17.3251
0.3631	48.0	25920	0.2735	17.2874
0.3529	49.0	26460	0.2734	17.0574
0.3478	50.0	27000	0.2746	16.9104
0.3401	51.0	27540	0.2720	16.9632
0.3365	52.0	28080	0.2731	16.9217
0.3276	53.0	28620	0.2752	16.8426
0.3261	54.0	29160	0.2734	16.5259
0.3158	55.0	29700	0.2795	16.4920
0.3174	56.0	30240	0.2741	16.3714
0.3127	57.0	30780	0.2775	16.3299
0.3059	58.0	31320	0.2803	16.2922
0.3064	59.0	31860	0.2770	16.2771
0.2959	60.0	32400	0.2836	16.2658
0.2974	61.0	32940	0.2835	16.2093
0.2870	62.0	33480	0.2881	16.3186
0.2823	63.0	34020	0.2885	16.0811
0.2784	64.0	34560	0.2782	16.0510
0.2757	65.0	35100	0.2877	16.0736
0.2725	66.0	35640	0.2828	15.7532
0.2642	67.0	36180	0.2895	15.7758
0.2614	68.0	36720	0.2900	15.9115
0.2570	69.0	37260	0.2999	15.6099
0.2567	70.0	37800	0.2920	15.6212
0.2490	71.0	38340	0.2967	15.5195
0.2517	72.0	38880	0.2997	15.6891
0.2439	73.0	39420	0.3015	15.3724
0.2416	74.0	39960	0.3125	15.5647
0.2340	75.0	40500	0.3050	15.2066
0.2300	76.0	41040	0.2965	15.2028
0.2303	77.0	41580	0.3028	15.2895
0.2252	78.0	42120	0.3087	15.0596
0.2206	79.0	42660	0.3105	15.2820
0.2196	80.0	43200	0.3055	15.0294
0.2175	81.0	43740	0.3166	15.0520
0.2168	82.0	44280	0.3153	15.3762
0.2132	83.0	44820	0.3157	15.1236

Framework versions

—Transformers 5.13.0
—Pytorch 2.6.0+cu124
—Datasets 3.6.0
—Tokenizers 0.22.2