<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->
kobart_8_5.6e-5_min30_lp5_sample_beams2
This model is a fine-tuned version of gogamza/kobart-base-v2 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 2.8283
- Rouge1: 35.819
- Rouge2: 12.1658
- Rougel: 23.3058
- Bleu1: 29.6395
- Bleu2: 16.8254
- Bleu3: 9.5014
- Bleu4: 5.168
- Gen Len: 49.8625
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: 5.6e-05
- train_batch_size: 8
- eval_batch_size: 128
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 5.0
Training results
| Training Loss | Epoch | Step | Validation Loss | Rouge1 | Rouge2 | Rougel | Bleu1 | Bleu2 | Bleu3 | Bleu4 | Gen Len |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 2.527 | 0.19 | 1000 | 3.0014 | 30.9895 | 9.5631 | 20.1782 | 25.4533 | 13.5291 | 7.1157 | 3.4483 | 50.2657 |
| 2.4214 | 0.38 | 2000 | 2.8814 | 32.3984 | 10.3443 | 21.2357 | 26.5661 | 14.5006 | 7.3531 | 3.5159 | 44.1538 |
| 2.3577 | 0.57 | 3000 | 2.8277 | 32.2306 | 10.5703 | 21.3959 | 26.4952 | 14.725 | 8.0596 | 4.2696 | 50.965 |
| 2.2606 | 0.76 | 4000 | 2.7749 | 33.0892 | 11.0109 | 21.4034 | 27.045 | 15.0797 | 8.2405 | 4.2337 | 48.1026 |
| 2.1508 | 0.94 | 5000 | 2.6841 | 33.1368 | 10.9332 | 21.9277 | 27.4808 | 15.2182 | 8.39 | 4.2468 | 46.0583 |
| 1.9467 | 1.13 | 6000 | 2.6994 | 33.2536 | 10.9192 | 21.851 | 26.7639 | 14.7669 | 8.1932 | 4.4866 | 42.7436 |
| 1.9267 | 1.32 | 7000 | 2.6743 | 35.335 | 12.5749 | 23.0923 | 29.4977 | 17.1053 | 9.9798 | 5.6851 | 54.2168 |
| 1.9402 | 1.51 | 8000 | 2.6549 | 34.7169 | 12.4365 | 22.8695 | 28.8948 | 16.8377 | 9.795 | 5.8984 | 53.8042 |
| 1.9457 | 1.7 | 9000 | 2.6198 | 34.1256 | 11.3508 | 22.7591 | 28.0771 | 15.6516 | 8.6198 | 4.5566 | 43.8252 |
| 1.9206 | 1.89 | 10000 | 2.6090 | 34.5521 | 12.0321 | 22.8654 | 28.268 | 16.2876 | 9.2697 | 4.9105 | 45.8205 |
| 1.6341 | 2.08 | 11000 | 2.6831 | 35.2143 | 12.748 | 23.2014 | 29.3413 | 17.2312 | 9.9515 | 5.5303 | 51.5338 |
| 1.6098 | 2.27 | 12000 | 2.6529 | 35.251 | 12.1877 | 23.3663 | 29.0609 | 16.6432 | 9.5808 | 5.2786 | 46.2378 |
| 1.6094 | 2.45 | 13000 | 2.6441 | 34.8683 | 12.0873 | 22.9699 | 28.9225 | 16.492 | 9.3451 | 5.1097 | 45.6131 |
| 1.6684 | 2.64 | 14000 | 2.6504 | 35.1897 | 12.0262 | 23.0832 | 28.948 | 16.4709 | 9.1994 | 5.0042 | 46.5245 |
| 1.6376 | 2.83 | 15000 | 2.6514 | 35.795 | 12.4779 | 23.2187 | 30.05 | 17.2789 | 9.984 | 5.4966 | 50.1119 |
| 1.3663 | 3.02 | 16000 | 2.7310 | 35.6544 | 12.109 | 23.3876 | 29.9268 | 16.945 | 9.4372 | 5.095 | 49.6317 |
| 1.3719 | 3.21 | 17000 | 2.7514 | 35.0663 | 11.8565 | 23.4224 | 28.8679 | 16.2846 | 9.3246 | 5.0154 | 45.3333 |
| 1.394 | 3.4 | 18000 | 2.7644 | 35.5883 | 12.2587 | 23.188 | 29.8503 | 17.0253 | 9.705 | 5.3253 | 47.4289 |
| 1.3615 | 3.59 | 19000 | 2.7535 | 35.3947 | 12.3879 | 23.355 | 29.4012 | 16.8473 | 9.6862 | 5.3268 | 48.7179 |
| 1.3544 | 3.78 | 20000 | 2.7480 | 35.7263 | 12.4434 | 23.6667 | 29.7146 | 17.0029 | 9.6018 | 5.2752 | 46.8834 |
| 1.3697 | 3.97 | 21000 | 2.7415 | 35.4189 | 12.1527 | 23.0022 | 29.6187 | 16.8477 | 9.5092 | 5.3766 | 50.3963 |
| 1.1718 | 4.15 | 22000 | 2.8251 | 35.0831 | 12.0809 | 22.8805 | 29.2252 | 16.5645 | 9.3818 | 5.241 | 46.7156 |
| 1.1955 | 4.34 | 23000 | 2.8158 | 35.7853 | 12.3885 | 23.821 | 29.7377 | 16.9635 | 9.7005 | 5.4376 | 47.5991 |
| 1.1795 | 4.53 | 24000 | 2.8265 | 35.4293 | 12.145 | 23.2029 | 29.6457 | 16.8228 | 9.7128 | 5.2525 | 49.5431 |
| 1.1835 | 4.72 | 25000 | 2.8254 | 35.499 | 11.9198 | 23.0859 | 29.4398 | 16.5715 | 9.2442 | 4.7663 | 47.8345 |
| 1.1644 | 4.91 | 26000 | 2.8283 | 35.819 | 12.1658 | 23.3058 | 29.6395 | 16.8254 | 9.5014 | 5.168 | 49.8625 |
Framework versions
- Transformers 4.25.1
- Pytorch 1.13.0+cu117
- Datasets 2.7.1
- Tokenizers 0.13.2
