metadata
library_name: peft
tags:
- alignment-handbook
- trl
- dpo
- generated_from_trainer
base_model: NbAiLab/nb-gpt-j-6B-v2
datasets:
- hugodk-sch/aftonposten_title_prefs
model-index:
- name: aftonposten-6b-align-scan
results: []
aftonposten-6b-align-scan
This model is a fine-tuned version of data/ap-gpt-j-6b-sft-qlora-04-08 on the hugodk-sch/aftonposten_title_prefs dataset. It achieves the following results on the evaluation set:
- Loss: 0.9279
- Rewards/chosen: -0.1421
- Rewards/rejected: -0.2171
- Rewards/accuracies: 0.5748
- Rewards/margins: 0.0750
- Logps/rejected: -38.0594
- Logps/chosen: -34.3898
- Logits/rejected: -2.1358
- Logits/chosen: -2.1406
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: 5e-06
- train_batch_size: 4
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- gradient_accumulation_steps: 2
- total_train_batch_size: 8
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 4
Training results
| Training Loss | Epoch | Step | Logits/chosen | Logits/rejected | Logps/chosen | Logps/rejected | Validation Loss | Rewards/accuracies | Rewards/chosen | Rewards/margins | Rewards/rejected |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.9521 | 0.26 | 100 | -2.2327 | -2.2279 | -34.0381 | -37.5401 | 0.9920 | 0.5361 | -0.0014 | 0.0080 | -0.0094 |
| 0.8515 | 0.52 | 200 | -2.2274 | -2.2226 | -34.0451 | -37.5730 | 0.9817 | 0.5594 | -0.0042 | 0.0184 | -0.0226 |
| 0.7473 | 0.78 | 300 | -2.2234 | -2.2186 | -34.0922 | -37.6165 | 0.9830 | 0.5390 | -0.0231 | 0.0169 | -0.0399 |
| 0.5492 | 1.04 | 400 | 0.9600 | -0.0653 | -0.1052 | 0.5565 | 0.0399 | -37.7797 | -34.1978 | -2.1852 | -2.1901 |
| 0.4504 | 1.3 | 500 | 0.9656 | -0.1403 | -0.1747 | 0.5685 | 0.0344 | -37.9533 | -34.3853 | -2.1770 | -2.1818 |
| 0.4511 | 1.56 | 600 | 0.9338 | -0.0835 | -0.1505 | 0.5951 | 0.0670 | -37.8930 | -34.2433 | -2.1565 | -2.1613 |
| 0.3805 | 1.82 | 700 | 0.9385 | -0.1061 | -0.1691 | 0.5511 | 0.0631 | -37.9394 | -34.2997 | -2.1442 | -2.1490 |
| 0.2038 | 2.08 | 800 | 0.9334 | -0.1274 | -0.1969 | 0.5689 | 0.0695 | -38.0088 | -34.3529 | -2.1367 | -2.1415 |
| 0.2332 | 2.34 | 900 | 0.9320 | -0.1353 | -0.2057 | 0.5718 | 0.0704 | -38.0309 | -34.3727 | -2.1352 | -2.1400 |
| 0.28 | 2.6 | 1000 | 0.9271 | -0.1454 | -0.2223 | 0.5714 | 0.0769 | -38.0723 | -34.3980 | -2.1345 | -2.1393 |
| 0.1953 | 2.86 | 1100 | 0.9399 | -0.1551 | -0.2188 | 0.5631 | 0.0637 | -38.0636 | -34.4223 | -2.1342 | -2.1390 |
| 0.2936 | 3.12 | 1200 | 0.9311 | -0.1412 | -0.2138 | 0.5864 | 0.0725 | -38.0510 | -34.3876 | -2.1359 | -2.1407 |
| 0.1526 | 3.38 | 1300 | 0.9307 | -0.1413 | -0.2150 | 0.5864 | 0.0737 | -38.0541 | -34.3877 | -2.1350 | -2.1398 |
| 0.1121 | 3.64 | 1400 | 0.9250 | -0.1377 | -0.2153 | 0.6038 | 0.0776 | -38.0548 | -34.3788 | -2.1356 | -2.1404 |
| 0.215 | 3.9 | 1500 | 0.9280 | -0.1384 | -0.2143 | 0.5806 | 0.0758 | -38.0523 | -34.3807 | -2.1358 | -2.1405 |
Framework versions
- PEFT 0.10.0
- Transformers 4.39.0.dev0
- Pytorch 2.1.2+cu121
- Datasets 2.14.6
- Tokenizers 0.15.1