wav2vec2-large-300m-colab-only-gn

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: 0.5274
• Wer: 0.5229

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: 0.0003
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
20.8148	0.45	25	13.5976	1.0
7.0188	0.9	50	5.5263	1.0
4.1285	1.35	75	3.6078	1.0
3.338	1.8	100	3.3217	1.0
3.2829	2.25	125	3.2781	1.0
3.272	2.7	150	3.2601	1.0
3.2224	3.15	175	3.2234	1.0
3.1949	3.6	200	3.1998	1.0
3.1846	4.05	225	3.1841	1.0
3.1615	4.5	250	3.1719	1.0
3.1367	4.95	275	3.1132	1.0
3.0111	5.41	300	2.9344	1.0
2.7786	5.86	325	2.5643	1.0
2.2106	6.31	350	1.8132	1.0
1.6365	6.76	375	1.4008	0.9982
1.178	7.21	400	1.0678	0.9845
0.8903	7.66	425	0.8744	0.9369
0.7429	8.11	450	0.7213	0.8752
0.5931	8.56	475	0.6681	0.8189
0.5592	9.01	500	0.6622	0.7895
0.4316	9.46	525	0.6177	0.7644
0.4098	9.91	550	0.5599	0.7874
0.3176	10.36	575	0.5649	0.7001
0.3142	10.81	600	0.5828	0.6867
0.3227	11.26	625	0.5505	0.6736
0.275	11.71	650	0.5432	0.6540
0.2783	12.16	675	0.5372	0.6462
0.2316	12.61	700	0.5078	0.6379
0.2281	13.06	725	0.5059	0.6161
0.2191	13.51	750	0.5175	0.5956
0.1911	13.96	775	0.5216	0.5929
0.1731	14.41	800	0.5069	0.5789
0.1743	14.86	825	0.5207	0.5971
0.1755	15.32	850	0.5436	0.6307
0.1568	15.77	875	0.5374	0.6001
0.1629	16.22	900	0.5429	0.6102
0.1418	16.67	925	0.5089	0.5762
0.136	17.12	950	0.5291	0.5878
0.1354	17.57	975	0.5381	0.5840
0.1351	18.02	1000	0.5511	0.5947
0.1252	18.47	1025	0.5204	0.5643
0.1215	18.92	1050	0.5385	0.5613
0.1188	19.37	1075	0.5063	0.5718
0.1209	19.82	1100	0.5211	0.5488
0.1091	20.27	1125	0.5245	0.5557
0.112	20.72	1150	0.4910	0.5587
0.102	21.17	1175	0.5192	0.5581
0.0947	21.62	1200	0.5500	0.5718
0.1066	22.07	1225	0.5288	0.5488
0.1011	22.52	1250	0.5180	0.5438
0.0974	22.97	1275	0.5089	0.5277
0.0926	23.42	1300	0.5222	0.5301
0.0871	23.87	1325	0.5135	0.5366
0.0808	24.32	1350	0.4990	0.5331
0.0739	24.77	1375	0.5281	0.5351
0.0841	25.23	1400	0.5321	0.5360
0.0743	25.68	1425	0.5508	0.5447
0.0809	26.13	1450	0.5228	0.5396
0.0631	26.58	1475	0.5284	0.5351
0.0788	27.03	1500	0.5250	0.5289
0.0754	27.48	1525	0.5204	0.5259
0.0663	27.93	1550	0.5275	0.5313
0.0645	28.38	1575	0.5288	0.5259
0.0729	28.83	1600	0.5268	0.5259
0.0656	29.28	1625	0.5277	0.5232
0.0703	29.73	1650	0.5274	0.5229

Framework versions

• Transformers 4.34.1
• Pytorch 2.0.1+cu118
• Datasets 2.14.6
• Tokenizers 0.14.1