results
This model is a fine-tuned version of microsoft/phi-1_5 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.0001
- Rewards/chosen: -7.5874
- Rewards/rejected: -24.0497
- Rewards/accuracies: 1.0
- Rewards/margins: 16.4623
- Logps/rejected: -274.3435
- Logps/chosen: -143.2090
- Logits/rejected: -1.8100
- Logits/chosen: -1.4786
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: 0.0005
- train_batch_size: 4
- eval_batch_size: 1
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 100
- training_steps: 1500
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.0417 | 0.07 | 100 | 0.0418 | -0.3892 | -8.0118 | 0.9792 | 7.6226 | -113.9640 | -71.2264 | 1.8258 | 1.7898 |
| 0.0221 | 0.15 | 200 | 0.0303 | -2.5657 | -10.9212 | 0.9896 | 8.3555 | -143.0585 | -92.9920 | 1.9704 | 2.1047 |
| 0.0107 | 0.22 | 300 | 0.0131 | -1.7388 | -11.6047 | 0.9965 | 9.8659 | -149.8935 | -84.7232 | 1.0731 | 0.9750 |
| 0.0204 | 0.29 | 400 | 0.0108 | -2.0131 | -11.9647 | 0.9965 | 9.9516 | -153.4932 | -87.4658 | 1.3610 | 1.6740 |
| 0.0067 | 0.36 | 500 | 0.0080 | -5.9488 | -19.6561 | 0.9974 | 13.7073 | -230.4076 | -126.8228 | -0.4464 | -0.2114 |
| 0.0 | 0.44 | 600 | 0.0047 | -5.6456 | -20.2381 | 0.9983 | 14.5924 | -236.2268 | -123.7909 | -0.4142 | -0.0244 |
| 0.0003 | 0.51 | 700 | 0.0018 | -7.2250 | -21.3351 | 0.9991 | 14.1101 | -247.1974 | -139.5853 | -0.3510 | -0.0203 |
| 0.0005 | 0.58 | 800 | 0.0008 | -7.2263 | -21.2475 | 0.9991 | 14.0211 | -246.3209 | -139.5981 | -0.8673 | -0.7010 |
| 0.0 | 0.66 | 900 | 0.0009 | -10.2371 | -26.0402 | 0.9991 | 15.8031 | -294.2486 | -169.7062 | -1.9784 | -1.7799 |
| 0.0 | 0.73 | 1000 | 0.0008 | -5.9544 | -22.0767 | 0.9991 | 16.1223 | -254.6137 | -126.8789 | -1.0623 | -0.6039 |
| 0.0 | 0.8 | 1100 | 0.0007 | -7.3374 | -23.8700 | 0.9991 | 16.5327 | -272.5467 | -140.7083 | -1.5517 | -1.1710 |
| 0.0 | 0.87 | 1200 | 0.0007 | -7.6398 | -24.1605 | 0.9991 | 16.5207 | -275.4509 | -143.7327 | -1.8124 | -1.4901 |
| 0.0 | 0.95 | 1300 | 0.0001 | -7.5920 | -24.0476 | 1.0 | 16.4556 | -274.3220 | -143.2550 | -1.8115 | -1.4816 |
| 0.0001 | 1.02 | 1400 | 0.0001 | -7.5872 | -24.0480 | 1.0 | 16.4608 | -274.3262 | -143.2065 | -1.8102 | -1.4791 |
| 0.0 | 1.09 | 1500 | 0.0001 | -7.5874 | -24.0497 | 1.0 | 16.4623 | -274.3435 | -143.2090 | -1.8100 | -1.4786 |
Framework versions
- Transformers 4.33.2
- Pytorch 2.0.1+cu118
- Datasets 2.14.5
- Tokenizers 0.13.3
Model tree for TrevorJS/mtg-phi-1_5-dpo-qlora
Base model
microsoft/phi-1_5
Finetuned
this model