llama-7b-SFT-qlora-eli5_DPO_ds_RM_top_2_1024_r_64_alpha_16
This model is a fine-tuned version of dhmeltzer/llama-7b-SFT_ds_eli5_1024_r_64_alpha_16_merged on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.6646
- Rewards/chosen: -0.0509
- Rewards/rejected: -0.1595
- Rewards/accuracies: 0.5955
- Rewards/margins: 0.1086
- Logps/rejected: -205.7926
- Logps/chosen: -209.8076
- Logits/rejected: 1.2037
- Logits/chosen: 1.2178
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6921 | 0.1 | 19 | 0.6816 | -0.1096 | -0.1495 | 0.5619 | 0.0399 | -205.6920 | -210.3942 | 1.2101 | 1.2240 |
| 0.6729 | 0.21 | 38 | 0.6800 | -0.3511 | -0.4714 | 0.5670 | 0.1204 | -208.9117 | -212.8093 | 1.1699 | 1.1829 |
| 0.6815 | 0.31 | 57 | 0.6718 | -0.1438 | -0.2304 | 0.5850 | 0.0866 | -206.5014 | -210.7368 | 1.1796 | 1.1924 |
| 0.6656 | 0.42 | 76 | 0.6670 | -0.1608 | -0.2728 | 0.6017 | 0.1120 | -206.9256 | -210.9071 | 1.1690 | 1.1824 |
| 0.6735 | 0.52 | 95 | 0.6656 | -0.0713 | -0.1981 | 0.5948 | 0.1268 | -206.1783 | -210.0114 | 1.1820 | 1.1939 |
| 0.6715 | 0.63 | 114 | 0.6672 | -0.0590 | -0.1724 | 0.5839 | 0.1134 | -205.9213 | -209.8885 | 1.2077 | 1.2215 |
| 0.6722 | 0.73 | 133 | 0.6659 | -0.0568 | -0.1635 | 0.5873 | 0.1067 | -205.8329 | -209.8666 | 1.2080 | 1.2218 |
| 0.6682 | 0.84 | 152 | 0.6646 | -0.0509 | -0.1595 | 0.5955 | 0.1086 | -205.7926 | -209.8076 | 1.2037 | 1.2178 |
| 0.673 | 0.94 | 171 | 0.6652 | -0.0532 | -0.1609 | 0.5960 | 0.1077 | -205.8064 | -209.8306 | 1.1965 | 1.2099 |
Framework versions
- Transformers 4.32.1
- Pytorch 2.0.1+cu118
- Datasets 2.14.4
- Tokenizers 0.13.3