metadata
base_model: ./data/zephyr-7b-sft-full
tags:
- alignment-handbook
- generated_from_trainer
datasets:
- HuggingFaceH4/ultrafeedback_binarized
model-index:
- name: zephyr-7b-dpo-full
results: []
zephyr-7b-dpo-full
This model is a fine-tuned version of ./data/zephyr-7b-sft-full on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:
- Loss: 0.5105
- Rewards/chosen: -1.7322
- Rewards/rejected: -3.3299
- Rewards/accuracies: 0.7619
- Rewards/margins: 1.5977
- Logps/rejected: -315.2173
- Logps/chosen: -359.3560
- Logits/rejected: -0.7333
- Logits/chosen: -0.7199
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-07
- train_batch_size: 8
- eval_batch_size: 4
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- total_train_batch_size: 64
- total_eval_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 3
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6517 | 0.1 | 100 | 0.6389 | -0.0070 | -0.1621 | 0.6905 | 0.1551 | -283.5396 | -342.1045 | -0.5321 | -0.4793 |
| 0.5605 | 0.21 | 200 | 0.5619 | -0.0146 | -0.6024 | 0.7381 | 0.5879 | -287.9430 | -342.1800 | -0.5264 | -0.4852 |
| 0.5581 | 0.31 | 300 | 0.5333 | -0.0290 | -0.8509 | 0.7540 | 0.8219 | -290.4272 | -342.3241 | -0.5108 | -0.4742 |
| 0.5467 | 0.41 | 400 | 0.5165 | -0.1986 | -1.1136 | 0.7698 | 0.9150 | -293.0540 | -344.0201 | -0.5404 | -0.5044 |
| 0.5223 | 0.52 | 500 | 0.5120 | -0.1374 | -1.1105 | 0.7659 | 0.9730 | -293.0233 | -343.4084 | -0.5315 | -0.4944 |
| 0.5265 | 0.62 | 600 | 0.5085 | -0.2099 | -1.2965 | 0.7698 | 1.0866 | -294.8834 | -344.1335 | -0.5350 | -0.4980 |
| 0.5342 | 0.72 | 700 | 0.4961 | -0.1152 | -1.1322 | 0.7738 | 1.0170 | -293.2408 | -343.1862 | -0.5509 | -0.5124 |
| 0.48 | 0.83 | 800 | 0.4913 | -0.1837 | -1.1984 | 0.7619 | 1.0148 | -293.9029 | -343.8708 | -0.5183 | -0.4760 |
| 0.517 | 0.93 | 900 | 0.4865 | -0.1696 | -1.2078 | 0.7659 | 1.0382 | -293.9965 | -343.7298 | -0.5289 | -0.4854 |
| 0.477 | 1.03 | 1000 | 0.4905 | -0.1084 | -1.2175 | 0.7619 | 1.1090 | -294.0931 | -343.1185 | -0.5469 | -0.5062 |
| 0.4033 | 1.14 | 1100 | 0.4870 | -0.1598 | -1.2266 | 0.7540 | 1.0668 | -294.1847 | -343.6326 | -0.5547 | -0.5138 |
| 0.3284 | 1.24 | 1200 | 0.4836 | -0.3432 | -1.5002 | 0.7817 | 1.1570 | -296.9207 | -345.4664 | -0.5812 | -0.5440 |
| 0.2574 | 1.34 | 1300 | 0.4861 | -0.5667 | -1.8467 | 0.7738 | 1.2801 | -300.3859 | -347.7009 | -0.5840 | -0.5523 |
| 0.2641 | 1.44 | 1400 | 0.4897 | -0.6824 | -1.9954 | 0.7698 | 1.3129 | -301.8724 | -348.8586 | -0.6308 | -0.6034 |
| 0.2424 | 1.55 | 1500 | 0.5010 | -0.8646 | -2.2932 | 0.7540 | 1.4286 | -304.8503 | -350.6802 | -0.6025 | -0.5800 |
| 0.2944 | 1.65 | 1600 | 0.4927 | -0.7608 | -2.1089 | 0.7659 | 1.3480 | -303.0073 | -349.6426 | -0.6171 | -0.5909 |
| 0.2958 | 1.75 | 1700 | 0.4913 | -0.8080 | -2.1126 | 0.7698 | 1.3046 | -303.0449 | -350.1146 | -0.6429 | -0.6156 |
| 0.2667 | 1.86 | 1800 | 0.4877 | -0.9185 | -2.2364 | 0.7619 | 1.3178 | -304.2823 | -351.2196 | -0.6212 | -0.5936 |
| 0.2494 | 1.96 | 1900 | 0.4853 | -0.8965 | -2.2705 | 0.75 | 1.3740 | -304.6238 | -350.9996 | -0.6262 | -0.6005 |
| 0.2631 | 2.06 | 2000 | 0.4869 | -0.7974 | -2.1804 | 0.7698 | 1.3830 | -303.7225 | -350.0081 | -0.6231 | -0.5974 |
| 0.1965 | 2.17 | 2100 | 0.4886 | -1.0005 | -2.3981 | 0.7540 | 1.3977 | -305.8999 | -352.0387 | -0.6557 | -0.6330 |
| 0.1711 | 2.27 | 2200 | 0.4910 | -1.1688 | -2.6422 | 0.7778 | 1.4734 | -308.3400 | -353.7221 | -0.6689 | -0.6486 |
| 0.1492 | 2.37 | 2300 | 0.5077 | -1.4306 | -3.0185 | 0.7778 | 1.5878 | -312.1035 | -356.3406 | -0.7016 | -0.6846 |
| 0.1448 | 2.48 | 2400 | 0.5113 | -1.6343 | -3.3087 | 0.7659 | 1.6744 | -315.0052 | -358.3771 | -0.7281 | -0.7164 |
| 0.1425 | 2.58 | 2500 | 0.5185 | -1.6767 | -3.4070 | 0.7698 | 1.7304 | -315.9888 | -358.8008 | -0.7207 | -0.7101 |
| 0.1661 | 2.68 | 2600 | 0.5144 | -1.6680 | -3.3881 | 0.7659 | 1.7201 | -315.7997 | -358.7144 | -0.7288 | -0.7184 |
| 0.1755 | 2.79 | 2700 | 0.5153 | -1.7546 | -3.3676 | 0.7619 | 1.6130 | -315.5941 | -359.5799 | -0.7388 | -0.7261 |
| 0.1677 | 2.89 | 2800 | 0.5120 | -1.7415 | -3.3279 | 0.7540 | 1.5863 | -315.1972 | -359.4494 | -0.7350 | -0.7219 |
| 0.1711 | 2.99 | 2900 | 0.5120 | -1.7362 | -3.3282 | 0.7619 | 1.5920 | -315.2005 | -359.3962 | -0.7329 | -0.7195 |
Framework versions
- Transformers 4.35.2
- Pytorch 2.1.1
- Datasets 2.14.7
- Tokenizers 0.14.1