b19-wav2vec2-large-xls-r-romansh-colab

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_13_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3604
• Wer: 0.2990

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: 0.0001
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
11.864	0.38	50	4.3744	1.0
3.6659	0.76	100	3.0489	1.0
2.9989	1.14	150	2.9634	1.0
2.9763	1.52	200	3.0247	1.0
2.954	1.9	250	2.9476	1.0
2.9795	2.29	300	3.0039	1.0
2.9296	2.67	350	2.9468	1.0
2.9904	3.05	400	2.9507	1.0
2.9245	3.43	450	2.9430	1.0
2.9721	3.81	500	2.9444	1.0
2.949	4.2	550	2.9392	1.0
2.9766	4.58	600	2.9383	1.0
2.9396	4.96	650	2.9393	1.0
2.9899	5.34	700	2.9373	1.0
2.9325	5.72	750	2.9331	1.0
2.9492	6.11	800	2.9287	1.0
2.9493	6.49	850	2.9379	1.0
2.9095	6.87	900	2.9258	1.0
2.9572	7.25	950	2.9377	1.0
2.9053	7.63	1000	2.9177	1.0
3.0002	8.02	1050	2.9122	1.0
2.8892	8.4	1100	2.8918	1.0
2.8927	8.78	1150	2.8707	1.0
2.8945	9.16	1200	2.8234	0.9991
2.7721	9.54	1250	2.4910	1.0
2.283	9.92	1300	1.5046	0.9991
1.6257	10.3	1350	1.0799	0.9998
1.2815	10.68	1400	0.8832	0.8158
1.1252	11.07	1450	0.7531	0.7189
0.9593	11.45	1500	0.6509	0.6309
0.8695	11.83	1550	0.6119	0.5836
0.8199	12.21	1600	0.5643	0.5680
0.7481	12.59	1650	0.5298	0.5340
0.7029	12.97	1700	0.5342	0.5254
0.6472	13.36	1750	0.5047	0.4856
0.6388	13.74	1800	0.4892	0.4844
0.5849	14.12	1850	0.4585	0.4548
0.5642	14.5	1900	0.4496	0.4497
0.546	14.88	1950	0.4274	0.4218
0.5345	15.27	2000	0.4218	0.4176
0.4925	15.65	2050	0.3989	0.3996
0.4794	16.03	2100	0.3887	0.3936
0.4528	16.41	2150	0.4039	0.3922
0.4542	16.79	2200	0.4135	0.3882
0.4408	17.17	2250	0.4018	0.3826
0.4144	17.56	2300	0.4072	0.3701
0.4048	17.94	2350	0.4026	0.3759
0.3751	18.32	2400	0.3774	0.3570
0.393	18.7	2450	0.3799	0.3654
0.3763	19.08	2500	0.3820	0.3484
0.3657	19.46	2550	0.3839	0.3500
0.3483	19.84	2600	0.3967	0.3505
0.3601	20.23	2650	0.3854	0.3524
0.3432	20.61	2700	0.3770	0.3428
0.3382	20.99	2750	0.3797	0.3337
0.3277	21.37	2800	0.3922	0.3414
0.3192	21.75	2850	0.3838	0.3479
0.3087	22.14	2900	0.3766	0.3337
0.3165	22.52	2950	0.3712	0.3242
0.3197	22.9	3000	0.3640	0.3207
0.317	23.28	3050	0.3630	0.3232
0.3123	23.66	3100	0.3777	0.3302
0.2804	24.05	3150	0.3640	0.3156
0.2768	24.43	3200	0.3598	0.3034
0.3038	24.81	3250	0.3627	0.3088
0.2716	25.19	3300	0.3571	0.3093
0.2911	25.57	3350	0.3619	0.3102
0.2903	25.95	3400	0.3697	0.3167
0.291	26.33	3450	0.3579	0.3093
0.2721	26.71	3500	0.3726	0.3128
0.2589	27.1	3550	0.3603	0.3020
0.2745	27.48	3600	0.3704	0.3074
0.2757	27.86	3650	0.3605	0.3018
0.2677	28.24	3700	0.3572	0.3034
0.2613	28.62	3750	0.3614	0.3014
0.2758	29.01	3800	0.3637	0.2979
0.2652	29.39	3850	0.3615	0.3000
0.2736	29.77	3900	0.3604	0.2990

Framework versions

• Transformers 4.26.0
• Pytorch 2.0.1+cu118
• Datasets 2.14.4
• Tokenizers 0.13.3