ASR-test-1

This model is a fine-tuned version of facebook/mms-1b-all on the HTV news dataset. It achieves the following results on the evaluation set:

• Loss: 0.6593
• Wer: 0.2797

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: 0.001
• train_batch_size: 32
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.8562	0.92	100	0.8316	0.4500
1.0777	1.83	200	0.6898	0.3899
0.98	2.75	300	0.6811	0.3740
0.8967	3.67	400	0.6332	0.3565
0.8965	4.59	500	0.6038	0.3517
0.8396	5.5	600	0.6040	0.3479
0.8137	6.42	700	0.5929	0.3408
0.8304	7.34	800	0.5911	0.3513
0.7894	8.26	900	0.6078	0.3357
0.7412	9.17	1000	0.6214	0.3230
0.7653	10.09	1100	0.5869	0.3444
0.7437	11.01	1200	0.5906	0.3213
0.7083	11.93	1300	0.5952	0.3139
0.7168	12.84	1400	0.5721	0.3267
0.7008	13.76	1500	0.5895	0.3177
0.6825	14.68	1600	0.5909	0.3098
0.6989	15.6	1700	0.5979	0.3673
0.6717	16.51	1800	0.5863	0.3077
0.6496	17.43	1900	0.5798	0.3043
0.6609	18.35	2000	0.5787	0.3555
0.628	19.27	2100	0.5889	0.3133
0.6322	20.18	2200	0.5913	0.3077
0.634	21.1	2300	0.5769	0.3193
0.6172	22.02	2400	0.5731	0.3005
0.6043	22.94	2500	0.5820	0.3075
0.6051	23.85	2600	0.5831	0.3435
0.5865	24.77	2700	0.5790	0.3029
0.5806	25.69	2800	0.5945	0.3053
0.5901	26.61	2900	0.5780	0.3126
0.5769	27.52	3000	0.5732	0.2963
0.5539	28.44	3100	0.5837	0.2950
0.5799	29.36	3200	0.5835	0.3178
0.5518	30.28	3300	0.5941	0.2943
0.549	31.19	3400	0.5960	0.2979
0.5612	32.11	3500	0.5747	0.3167
0.5411	33.03	3600	0.5855	0.2978
0.536	33.94	3700	0.5720	0.2944
0.5329	34.86	3800	0.5998	0.3186
0.5185	35.78	3900	0.5936	0.2884
0.5186	36.7	4000	0.5773	0.2901
0.5027	37.61	4100	0.5969	0.3264
0.52	38.53	4200	0.6184	0.2939
0.4992	39.45	4300	0.5887	0.2943
0.5064	40.37	4400	0.5814	0.2966
0.4928	41.28	4500	0.6128	0.2902
0.508	42.2	4600	0.5943	0.2923
0.4887	43.12	4700	0.6100	0.3039
0.4872	44.04	4800	0.6044	0.2875
0.4711	44.95	4900	0.5961	0.2974
0.4813	45.87	5000	0.6022	0.2945
0.4818	46.79	5100	0.6199	0.2898
0.4492	47.71	5200	0.6161	0.2943
0.4715	48.62	5300	0.6038	0.2838
0.4601	49.54	5400	0.6223	0.2829
0.4432	50.46	5500	0.6058	0.2965
0.4419	51.38	5600	0.6134	0.2917
0.4564	52.29	5700	0.6124	0.2857
0.4349	53.21	5800	0.6229	0.2877
0.4358	54.13	5900	0.6095	0.2898
0.4432	55.05	6000	0.6365	0.2881
0.4277	55.96	6100	0.6169	0.2870
0.4397	56.88	6200	0.6174	0.2849
0.4245	57.8	6300	0.6340	0.2858
0.4203	58.72	6400	0.6321	0.2909
0.4112	59.63	6500	0.6243	0.2866
0.4244	60.55	6600	0.6318	0.2775
0.4119	61.47	6700	0.6215	0.2798
0.403	62.39	6800	0.6213	0.2829
0.4158	63.3	6900	0.6451	0.2795
0.3997	64.22	7000	0.6317	0.2854
0.4006	65.14	7100	0.6329	0.2846
0.4051	66.06	7200	0.6318	0.2834
0.3953	66.97	7300	0.6442	0.2855
0.4119	67.89	7400	0.6345	0.2893
0.3976	68.81	7500	0.6361	0.2798
0.3965	69.72	7600	0.6355	0.2853
0.3957	70.64	7700	0.6457	0.2814
0.3837	71.56	7800	0.6396	0.2855
0.3893	72.48	7900	0.6424	0.2842
0.3816	73.39	8000	0.6496	0.2778
0.3855	74.31	8100	0.6427	0.2881
0.3767	75.23	8200	0.6394	0.2858
0.3747	76.15	8300	0.6513	0.2844
0.3829	77.06	8400	0.6602	0.2775
0.3721	77.98	8500	0.6427	0.2825
0.3708	78.9	8600	0.6507	0.2847
0.3767	79.82	8700	0.6518	0.2816
0.3655	80.73	8800	0.6597	0.2802
0.3614	81.65	8900	0.6542	0.2781
0.3629	82.57	9000	0.6520	0.2782
0.3621	83.49	9100	0.6501	0.2797
0.3616	84.4	9200	0.6528	0.2777
0.3519	85.32	9300	0.6549	0.2798
0.3572	86.24	9400	0.6541	0.2789
0.3585	87.16	9500	0.6497	0.2778
0.3531	88.07	9600	0.6523	0.2781
0.3586	88.99	9700	0.6578	0.2789
0.3463	89.91	9800	0.6565	0.2816
0.3508	90.83	9900	0.6559	0.2797
0.3513	91.74	10000	0.6611	0.2794
0.3425	92.66	10100	0.6538	0.2804
0.3596	93.58	10200	0.6639	0.2808
0.3632	94.5	10300	0.6561	0.2789
0.348	95.41	10400	0.6556	0.2786
0.3514	96.33	10500	0.6575	0.2791
0.3499	97.25	10600	0.6573	0.2795
0.3353	98.17	10700	0.6589	0.2797
0.3468	99.08	10800	0.6589	0.2799
0.3571	100.0	10900	0.6593	0.2797

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.0