metadata
library_name: peft
tags:
- trl
- dpo
- alignment-handbook
- generated_from_trainer
base_model: NbAiLab/nb-gpt-j-6B-v2
model-index:
- name: aftonposten-6b-align-scan
results: []
aftonposten-6b-align-scan
This model is a fine-tuned version of NbAiLab/nb-gpt-j-6B-v2 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.3543
- Rewards/chosen: 0.1332
- Rewards/rejected: 0.1192
- Rewards/accuracies: 0.5486
- Rewards/margins: 0.0139
- Logps/rejected: -37.3842
- Logps/chosen: -33.8866
- Logits/rejected: -2.2421
- Logits/chosen: -2.2469
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.3038 | 0.26 | 100 | -2.2372 | -2.2324 | -34.0128 | -37.5115 | 0.3512 | 0.5424 | 0.0196 | 0.0150 | 0.0046 |
| 0.3157 | 0.52 | 200 | -2.2371 | -2.2322 | -34.0181 | -37.5184 | 0.3716 | 0.5245 | 0.0148 | 0.0164 | -0.0016 |
| 0.2156 | 0.78 | 300 | -2.2364 | -2.2316 | -34.0143 | -37.4970 | 0.3845 | 0.4934 | 0.0182 | 0.0005 | 0.0177 |
| 0.4084 | 1.04 | 400 | 0.4059 | 0.0705 | 0.0718 | 0.5066 | -0.0013 | -37.4369 | -33.9562 | -2.2400 | -2.2448 |
| 0.2788 | 1.3 | 500 | 0.3866 | 0.0701 | 0.0576 | 0.5191 | 0.0125 | -37.4526 | -33.9566 | -2.2356 | -2.2405 |
| 0.3874 | 1.56 | 600 | 0.4265 | 0.0711 | 0.0890 | 0.4726 | -0.0180 | -37.4177 | -33.9556 | -2.2421 | -2.2470 |
| 0.2695 | 1.82 | 700 | 0.4028 | 0.0816 | 0.0876 | 0.5079 | -0.0060 | -37.4193 | -33.9439 | -2.2429 | -2.2478 |
| 0.1725 | 2.08 | 800 | 0.4083 | 0.0967 | 0.1077 | 0.4821 | -0.0110 | -37.3970 | -33.9271 | -2.2415 | -2.2463 |
| 0.2502 | 2.34 | 900 | 0.4099 | 0.1154 | 0.1311 | 0.4900 | -0.0157 | -37.3709 | -33.9064 | -2.2438 | -2.2487 |
| 0.1529 | 2.6 | 1000 | 0.3879 | 0.1222 | 0.1257 | 0.5216 | -0.0034 | -37.3770 | -33.8988 | -2.2428 | -2.2477 |
| 0.1583 | 2.86 | 1100 | 0.3968 | 0.1193 | 0.1250 | 0.4875 | -0.0057 | -37.3777 | -33.9020 | -2.2433 | -2.2482 |
| 0.113 | 3.12 | 1200 | 0.3849 | 0.1137 | 0.1163 | 0.4784 | -0.0025 | -37.3874 | -33.9082 | -2.2421 | -2.2470 |
| 0.0937 | 3.38 | 1300 | 0.3738 | 0.1235 | 0.1177 | 0.5046 | 0.0058 | -37.3859 | -33.8973 | -2.2423 | -2.2472 |
| 0.0815 | 3.64 | 1400 | 0.3595 | 0.1338 | 0.1197 | 0.5224 | 0.0141 | -37.3836 | -33.8859 | -2.2427 | -2.2476 |
| 0.0757 | 3.9 | 1500 | 0.3543 | 0.1332 | 0.1192 | 0.5486 | 0.0139 | -37.3842 | -33.8866 | -2.2421 | -2.2469 |
Framework versions
- PEFT 0.10.0
- Transformers 4.39.0.dev0
- Pytorch 2.1.2+cu121
- Datasets 2.14.6
- Tokenizers 0.15.1