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: 0.4994
- Rewards/chosen: 0.0025
- Rewards/rejected: -0.0003
- Rewards/accuracies: 0.6038
- Rewards/margins: 0.0028
- Logps/rejected: -37.5434
- Logps/chosen: -33.7834
- Logits/rejected: -2.0909
- Logits/chosen: -2.0955
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.4997 | 0.26 | 100 | -2.2320 | -2.2272 | -33.8530 | -37.3521 | 0.5000 | 0.5183 | 0.0018 | 0.0002 | 0.0016 |
| 0.4992 | 0.52 | 200 | -2.2288 | -2.2240 | -33.7422 | -37.2812 | 0.4999 | 0.5748 | 0.0029 | 0.0006 | 0.0024 |
| 0.4982 | 0.78 | 300 | -2.2289 | -2.2241 | -33.7545 | -37.2741 | 0.4999 | 0.5042 | 0.0028 | 0.0004 | 0.0024 |
| 0.4967 | 1.04 | 400 | 0.4999 | 0.0036 | 0.0030 | 0.5623 | 0.0006 | -37.2152 | -33.6722 | -2.2047 | -2.2095 |
| 0.4954 | 1.3 | 500 | 0.4998 | 0.0042 | 0.0033 | 0.5772 | 0.0009 | -37.1891 | -33.6191 | -2.1759 | -2.1807 |
| 0.4964 | 1.56 | 600 | 0.4996 | 0.0044 | 0.0026 | 0.5685 | 0.0018 | -37.2574 | -33.5925 | -2.1522 | -2.1569 |
| 0.4942 | 1.82 | 700 | 0.4996 | 0.0034 | 0.0018 | 0.5797 | 0.0016 | -37.3406 | -33.6981 | -2.1370 | -2.1417 |
| 0.4926 | 2.08 | 800 | 0.4995 | 0.0037 | 0.0013 | 0.5947 | 0.0024 | -37.3864 | -33.6678 | -2.1276 | -2.1323 |
| 0.491 | 2.34 | 900 | 0.4995 | 0.0031 | 0.0006 | 0.5739 | 0.0025 | -37.4562 | -33.7214 | -2.1107 | -2.1154 |
| 0.4941 | 2.6 | 1000 | 0.4995 | 0.0028 | 0.0001 | 0.6059 | 0.0027 | -37.5025 | -33.7503 | -2.0960 | -2.1006 |
| 0.4943 | 2.86 | 1100 | 0.4995 | 0.0025 | -0.0002 | 0.5826 | 0.0028 | -37.5382 | -33.7800 | -2.0918 | -2.0965 |
| 0.4911 | 3.12 | 1200 | 0.4995 | 0.0026 | -0.0001 | 0.6009 | 0.0027 | -37.5257 | -33.7754 | -2.0913 | -2.0959 |
| 0.4906 | 3.38 | 1300 | 0.4995 | 0.0025 | -0.0002 | 0.6034 | 0.0027 | -37.5366 | -33.7817 | -2.0909 | -2.0955 |
| 0.4928 | 3.64 | 1400 | 0.4995 | 0.0025 | -0.0003 | 0.5922 | 0.0028 | -37.5423 | -33.7843 | -2.0903 | -2.0950 |
| 0.4912 | 3.9 | 1500 | 0.4995 | 0.0025 | -0.0003 | 0.6034 | 0.0028 | -37.5427 | -33.7835 | -2.0908 | -2.0954 |
Framework versions
- PEFT 0.8.2
- Transformers 4.37.2
- Pytorch 2.1.2+cu121
- Datasets 2.17.0
- Tokenizers 0.15.1