wav2vec2-base-intent-classification-ori-f1
This model is a fine-tuned version of facebook/wav2vec2-base on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.4353
- F1: 0.875
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: 3e-05
- train_batch_size: 1
- eval_batch_size: 1
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 4
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 45
Training results
Training Loss | Epoch | Step | Validation Loss | F1 |
---|---|---|---|---|
2.19 | 1.0 | 28 | 2.1733 | 0.2708 |
2.1205 | 2.0 | 56 | 2.1125 | 0.2708 |
2.0965 | 3.0 | 84 | 2.0543 | 0.2708 |
1.9694 | 4.0 | 112 | 1.9125 | 0.2917 |
1.9091 | 5.0 | 140 | 1.8455 | 0.3542 |
1.8399 | 6.0 | 168 | 1.7895 | 0.3958 |
1.8424 | 7.0 | 196 | 1.8828 | 0.3125 |
1.5475 | 8.0 | 224 | 1.4255 | 0.5208 |
1.2653 | 9.0 | 252 | 1.3953 | 0.5417 |
1.1465 | 10.0 | 280 | 1.3501 | 0.5417 |
1.281 | 11.0 | 308 | 1.2800 | 0.5417 |
1.0996 | 12.0 | 336 | 1.2797 | 0.6042 |
1.1288 | 13.0 | 364 | 1.1341 | 0.6667 |
0.8577 | 14.0 | 392 | 1.0104 | 0.7083 |
0.8047 | 15.0 | 420 | 1.0906 | 0.6667 |
0.7098 | 16.0 | 448 | 0.9710 | 0.7917 |
0.5407 | 17.0 | 476 | 0.9363 | 0.7708 |
0.4634 | 18.0 | 504 | 0.8283 | 0.75 |
0.4368 | 19.0 | 532 | 0.7587 | 0.7708 |
0.2818 | 20.0 | 560 | 0.6551 | 0.8333 |
0.1951 | 21.0 | 588 | 0.5865 | 0.8333 |
0.1456 | 22.0 | 616 | 0.7378 | 0.7917 |
0.1269 | 23.0 | 644 | 0.6327 | 0.8333 |
0.0801 | 24.0 | 672 | 0.6896 | 0.8333 |
0.0723 | 25.0 | 700 | 0.7179 | 0.8333 |
0.0626 | 26.0 | 728 | 1.0643 | 0.7708 |
0.0434 | 27.0 | 756 | 0.4353 | 0.875 |
0.0499 | 28.0 | 784 | 0.6656 | 0.8333 |
0.0396 | 29.0 | 812 | 0.6788 | 0.8333 |
0.0352 | 30.0 | 840 | 0.8139 | 0.8333 |
0.0348 | 31.0 | 868 | 0.8745 | 0.8125 |
0.0313 | 32.0 | 896 | 0.8693 | 0.8125 |
0.0269 | 33.0 | 924 | 0.9393 | 0.8125 |
0.0242 | 34.0 | 952 | 0.9351 | 0.8333 |
0.0217 | 35.0 | 980 | 0.9406 | 0.8333 |
0.0234 | 36.0 | 1008 | 0.9464 | 0.8333 |
0.0219 | 37.0 | 1036 | 0.9507 | 0.8333 |
0.0215 | 38.0 | 1064 | 0.9471 | 0.8333 |
0.0206 | 39.0 | 1092 | 0.9260 | 0.8333 |
0.0229 | 40.0 | 1120 | 0.9420 | 0.8333 |
0.0216 | 41.0 | 1148 | 0.9570 | 0.8333 |
0.0227 | 42.0 | 1176 | 0.9573 | 0.8333 |
0.0208 | 43.0 | 1204 | 0.9609 | 0.8333 |
0.0201 | 44.0 | 1232 | 0.9617 | 0.8333 |
0.0208 | 45.0 | 1260 | 0.9620 | 0.8333 |
Framework versions
- Transformers 4.20.1
- Pytorch 1.11.0
- Datasets 2.1.0
- Tokenizers 0.12.1
- Downloads last month
- 10
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.