german-jeopardy-longt5-large-128
This model is a fine-tuned version of google/long-t5-tglobal-large on the lmqg/qg_dequad dataset. It achieves the following results on the evaluation set:
- Loss: 2.6149
- Brevity Penalty: 0.9386
- System Length: 19554
- Reference Length: 20793
- ROUGE-1: 28.96
- ROUGE-2: 11.91
- ROUGE-L: 27.92
- ROUGE-Lsum: 27.91
- Exact Match: 0.95
- BLEU: 6.99
- F1: 28.39
Model description
See google/long-t5-tglobal-large for more information about the
model architecture.
The model was trained on a single NVIDIA RTX 3090 GPU with 24GB of VRAM.
Intended uses & limitations
This model can be used for question generation on German text.
Training and evaluation data
See lmqg/qg_dequad.
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0001
- train_batch_size: 2
- eval_batch_size: 2
- seed: 7
- gradient_accumulation_steps: 64
- total_train_batch_size: 128
- optimizer: Adafactor
- lr_scheduler_type: constant
- num_epochs: 20
Training results
Training Loss | Epoch | Step | Validation Loss | Counts 1 | Counts 2 | Counts 3 | Counts 4 | Totals 1 | Totals 2 | Totals 3 | Totals 4 | Precisions 1 | Precisions 2 | Precisions 3 | Precisions 4 | Brevity Penalty | System Length | Reference Length | ROUGE-1 | ROUGE-2 | ROUGE-L | ROUGE-Lsum | Exact Match | BLEU | Mean Generated Length | F1 |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
7.5882 | 0.99 | 72 | 5.6823 | 3993 | 105 | 0 | 0 | 14790 | 12586 | 10382 | 8178 | 26.998 | 0.8343 | 0.0048 | 0.0031 | 0.6461 | 14790 | 21250 | 0.1101 | 0.0077 | 0.1078 | 0.1076 | 0.0 | 0.0872 | 9.7105 | 0.1155 |
5.2903 | 1.99 | 145 | 4.8721 | 3827 | 229 | 32 | 0 | 18894 | 16690 | 14486 | 12282 | 20.2551 | 1.3721 | 0.2209 | 0.0041 | 0.8828 | 18894 | 21250 | 0.0924 | 0.015 | 0.091 | 0.0909 | 0.0 | 0.351 | 16.7005 | 0.0964 |
4.6636 | 3.0 | 218 | 4.2806 | 3638 | 174 | 21 | 0 | 15268 | 13064 | 10860 | 8656 | 23.8276 | 1.3319 | 0.1934 | 0.0058 | 0.6758 | 15268 | 21250 | 0.0884 | 0.012 | 0.0876 | 0.0874 | 0.0 | 0.2933 | 8.9197 | 0.0925 |
4.2229 | 4.0 | 291 | 3.9210 | 4274 | 240 | 24 | 0 | 29308 | 27104 | 24900 | 22696 | 14.583 | 0.8855 | 0.0964 | 0.0022 | 1.0 | 29308 | 21250 | 0.0894 | 0.0109 | 0.0849 | 0.0849 | 0.0 | 0.2288 | 24.7015 | 0.1023 |
3.9434 | 4.99 | 363 | 3.6907 | 3652 | 218 | 35 | 1 | 16442 | 14238 | 12034 | 9830 | 22.2114 | 1.5311 | 0.2908 | 0.0102 | 0.7465 | 16442 | 21250 | 0.0856 | 0.0141 | 0.0843 | 0.0842 | 0.0 | 0.4204 | 12.3049 | 0.0898 |
3.6152 | 5.99 | 436 | 3.4603 | 4103 | 341 | 77 | 11 | 20581 | 18377 | 16173 | 13969 | 19.9359 | 1.8556 | 0.4761 | 0.0787 | 0.968 | 20581 | 21250 | 0.107 | 0.019 | 0.1023 | 0.1024 | 0.0 | 1.0505 | 14.3607 | 0.112 |
3.3814 | 7.0 | 509 | 3.2883 | 4342 | 675 | 218 | 43 | 17763 | 15559 | 13355 | 11151 | 24.4441 | 4.3383 | 1.6323 | 0.3856 | 0.8218 | 17763 | 21250 | 0.1264 | 0.0353 | 0.1234 | 0.1234 | 0.0005 | 2.3489 | 10.2418 | 0.1308 |
3.1711 | 8.0 | 582 | 3.0988 | 4820 | 856 | 246 | 44 | 19759 | 17555 | 15351 | 13147 | 24.3939 | 4.8761 | 1.6025 | 0.3347 | 0.9273 | 19759 | 21250 | 0.1503 | 0.0465 | 0.1455 | 0.1457 | 0.0005 | 2.6207 | 14.3249 | 0.1547 |
3.0147 | 8.99 | 654 | 2.9540 | 5167 | 1066 | 321 | 76 | 18725 | 16521 | 14317 | 12113 | 27.5941 | 6.4524 | 2.2421 | 0.6274 | 0.8739 | 18725 | 21250 | 0.1773 | 0.0588 | 0.1721 | 0.1721 | 0.0018 | 3.4764 | 14.3067 | 0.1816 |
2.7829 | 9.99 | 727 | 2.8288 | 5625 | 1267 | 420 | 124 | 17327 | 15123 | 12919 | 10715 | 32.4638 | 8.378 | 3.251 | 1.1573 | 0.7974 | 17327 | 21250 | 0.2127 | 0.0741 | 0.2067 | 0.2065 | 0.0045 | 4.5099 | 12.9741 | 0.2159 |
2.6093 | 10.99 | 800 | 2.7177 | 6005 | 1469 | 528 | 181 | 18625 | 16421 | 14217 | 12013 | 32.2416 | 8.9459 | 3.7139 | 1.5067 | 0.8685 | 18625 | 21250 | 0.229 | 0.0827 | 0.2215 | 0.2213 | 0.0064 | 5.5051 | 14.4791 | 0.231 |
2.453 | 12.0 | 873 | 2.5914 | 6396 | 1744 | 664 | 246 | 18307 | 16103 | 13899 | 11695 | 34.9375 | 10.8303 | 4.7773 | 2.1035 | 0.8515 | 18307 | 21250 | 0.2553 | 0.0998 | 0.2479 | 0.2478 | 0.0059 | 6.6865 | 13.7142 | 0.2565 |
2.3329 | 12.99 | 945 | 2.4993 | 6673 | 1888 | 741 | 291 | 18451 | 16247 | 14043 | 11839 | 36.1661 | 11.6206 | 5.2767 | 2.458 | 0.8592 | 18451 | 21250 | 0.2747 | 0.1114 | 0.2652 | 0.2652 | 0.0091 | 7.383 | 14.1751 | 0.2749 |
2.1663 | 13.99 | 1018 | 2.4196 | 6953 | 2052 | 834 | 337 | 18531 | 16327 | 14123 | 11919 | 37.5209 | 12.5681 | 5.9053 | 2.8274 | 0.8635 | 18531 | 21250 | 0.2886 | 0.1215 | 0.2773 | 0.277 | 0.0082 | 8.1343 | 14.6783 | 0.2889 |
2.0422 | 14.99 | 1091 | 2.3703 | 6968 | 2089 | 862 | 365 | 17984 | 15780 | 13576 | 11372 | 38.7456 | 13.2383 | 6.3494 | 3.2096 | 0.8339 | 17984 | 21250 | 0.2961 | 0.1268 | 0.2858 | 0.2857 | 0.0113 | 8.4322 | 13.6987 | 0.2951 |
1.9245 | 16.0 | 1164 | 2.3217 | 7500 | 2353 | 999 | 446 | 19017 | 16813 | 14609 | 12405 | 39.4384 | 13.9951 | 6.8383 | 3.5953 | 0.8892 | 19017 | 21250 | 0.3149 | 0.1407 | 0.3017 | 0.3017 | 0.0132 | 9.5973 | 14.77 | 0.314 |
1.8216 | 17.0 | 1237 | 2.2705 | 7444 | 2357 | 1044 | 488 | 18219 | 16015 | 13811 | 11607 | 40.8584 | 14.7175 | 7.5592 | 4.2044 | 0.8467 | 18219 | 21250 | 0.3201 | 0.1437 | 0.3081 | 0.3077 | 0.0132 | 9.9557 | 13.8031 | 0.3181 |
1.7503 | 17.99 | 1309 | 2.2386 | 7571 | 2487 | 1114 | 515 | 18275 | 16071 | 13867 | 11663 | 41.4282 | 15.4751 | 8.0335 | 4.4157 | 0.8498 | 18275 | 21250 | 0.3289 | 0.1512 | 0.3153 | 0.3151 | 0.0145 | 10.4354 | 13.9106 | 0.3265 |
1.6342 | 18.99 | 1382 | 2.2183 | 7697 | 2536 | 1155 | 537 | 18129 | 15925 | 13721 | 11517 | 42.4568 | 15.9246 | 8.4178 | 4.6627 | 0.8418 | 18129 | 21250 | 0.3342 | 0.1559 | 0.3224 | 0.3222 | 0.0177 | 10.7447 | 13.8494 | 0.3313 |
1.5474 | 19.79 | 1440 | 2.1956 | 7879 | 2632 | 1187 | 570 | 18815 | 16611 | 14407 | 12203 | 41.8762 | 15.8449 | 8.2391 | 4.671 | 0.8786 | 18815 | 21250 | 0.3398 | 0.1607 | 0.326 | 0.326 | 0.0177 | 11.1066 | 14.5136 | 0.3375 |
Framework versions
- Transformers 4.32.1
- Pytorch 2.1.0
- Datasets 2.12.0
- Tokenizers 0.13.3
- Downloads last month
- 11
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.
Dataset used to train GiantTreeG/german-jeopardy-longt5-large-128
Evaluation results
- BLEU-4 on lmqg/qg_dequadself-reported6.990
- F1 on lmqg/qg_dequadself-reported28.390
- ROUGE-1 on lmqg/qg_dequadself-reported28.960
- ROUGE-2 on lmqg/qg_dequadself-reported11.910
- ROUGE-L on lmqg/qg_dequadself-reported27.920
- ROUGE-Lsum on lmqg/qg_dequadself-reported27.910
- Exact Match on lmqg/qg_dequadself-reported0.950