gopdataset_phonome_base_add_transformer

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3081
• Cer: 0.1141

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: 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: 1000
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
6.7266	0.84	100	3.4268	0.9750
3.258	1.68	200	3.2266	0.7902
2.5421	2.52	300	1.1589	0.5124
1.0681	3.36	400	0.4367	0.1676
0.7192	4.2	500	0.4418	0.1658
0.5793	5.04	600	0.3079	0.1331
0.5329	5.88	700	0.3078	0.1287
0.4988	6.72	800	0.3051	0.1251
0.4455	7.56	900	0.2843	0.1206
0.4271	8.4	1000	0.2865	0.1234
0.4027	9.24	1100	0.2996	0.1214
0.3939	10.08	1200	0.2874	0.1199
0.3633	10.92	1300	0.2777	0.1237
0.3482	11.76	1400	0.2648	0.1171
0.3267	12.61	1500	0.2737	0.1174
0.3334	13.45	1600	0.2812	0.1176
0.3145	14.29	1700	0.2709	0.1163
0.2921	15.13	1800	0.2689	0.1153
0.2939	15.97	1900	0.2757	0.1153
0.2681	16.81	2000	0.2785	0.1161
0.2691	17.65	2100	0.2955	0.1196
0.2627	18.49	2200	0.2922	0.1174
0.2519	19.33	2300	0.2820	0.1148
0.2391	20.17	2400	0.3038	0.1190
0.2393	21.01	2500	0.2873	0.1162
0.2324	21.85	2600	0.2903	0.1148
0.2217	22.69	2700	0.3018	0.1167
0.2156	23.53	2800	0.3033	0.1153
0.2039	24.37	2900	0.2975	0.1147
0.2018	25.21	3000	0.3055	0.1159
0.1996	26.05	3100	0.3035	0.1151
0.2013	26.89	3200	0.3032	0.1153
0.2002	27.73	3300	0.3029	0.1146
0.196	28.57	3400	0.3118	0.1157
0.2047	29.41	3500	0.3081	0.1141

Framework versions

• Transformers 4.17.0
• Pytorch 2.4.0
• Datasets 1.18.3
• Tokenizers 0.20.3