llama-7b-SFT-qlora-eli5-wiki_DPO_ds_RM_contrast_1024_r_64_alpha_16
This model is a fine-tuned version of dhmeltzer/llama-7b-SFT_eli5_wiki65k_1024_r_64_alpha_16_merged on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.6234
- Rewards/chosen: 0.0858
- Rewards/rejected: -0.1898
- Rewards/accuracies: 0.6574
- Rewards/margins: 0.2756
- Logps/rejected: -198.1188
- Logps/chosen: -205.4868
- Logits/rejected: 0.7931
- Logits/chosen: 0.8315
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: 0.0002
- train_batch_size: 32
- eval_batch_size: 32
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 128
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.03
- num_epochs: 1
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6867 | 0.1 | 19 | 0.6390 | 0.0633 | -0.1318 | 0.6451 | 0.1951 | -197.8286 | -205.5991 | 0.7774 | 0.8133 |
| 0.6727 | 0.21 | 38 | 0.6384 | 0.0354 | -0.2285 | 0.6529 | 0.2639 | -198.3123 | -205.7386 | 0.8054 | 0.8432 |
| 0.6577 | 0.31 | 57 | 0.6391 | -0.0114 | -0.2258 | 0.6406 | 0.2145 | -198.2988 | -205.9725 | 0.7954 | 0.8346 |
| 0.6609 | 0.42 | 76 | 0.6344 | -0.3737 | -0.6175 | 0.6417 | 0.2438 | -200.2571 | -207.7841 | 0.7818 | 0.8194 |
| 0.6536 | 0.52 | 95 | 0.6285 | -0.1130 | -0.3816 | 0.6652 | 0.2687 | -199.0778 | -206.4805 | 0.7958 | 0.8350 |
| 0.654 | 0.62 | 114 | 0.6342 | 0.0007 | -0.2311 | 0.6484 | 0.2318 | -198.3250 | -205.9122 | 0.7917 | 0.8303 |
| 0.6435 | 0.73 | 133 | 0.6258 | 0.0462 | -0.2234 | 0.6562 | 0.2696 | -198.2865 | -205.6845 | 0.7949 | 0.8332 |
| 0.6508 | 0.83 | 152 | 0.6234 | 0.0858 | -0.1898 | 0.6574 | 0.2756 | -198.1188 | -205.4868 | 0.7931 | 0.8315 |
| 0.6361 | 0.94 | 171 | 0.6269 | 0.1007 | -0.1655 | 0.6618 | 0.2662 | -197.9971 | -205.4121 | 0.7975 | 0.8353 |
Framework versions
- Transformers 4.32.1
- Pytorch 2.0.1+cu118
- Datasets 2.14.4
- Tokenizers 0.13.3