metadata
license: mit
base_model: EleutherAI/gpt-neo-125m
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: gpt-neo-125m_hh_reward
results: []
gpt-neo-125m_hh_reward
This model is a DPO fine-tuned version of EleutherAI/gpt-neo-125m on Anthropics HH dataset. It achieves the following results on the evaluation set:
- Loss: 0.7764
- Rewards/chosen: -1.0726
- Rewards/rejected: -1.1588
- Rewards/accuracies: 0.5592
- Rewards/margins: 0.0861
- Logps/rejected: -127.6699
- Logps/chosen: -107.5507
- Logits/rejected: -15.9221
- Logits/chosen: -15.7406
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: 1e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 150
- training_steps: 20050
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.7685 | 0.1 | 2000 | 0.7585 | -0.6757 | -0.7643 | 0.5066 | 0.0886 | -126.6837 | -106.5583 | -16.3107 | -16.1234 |
| 0.795 | 0.2 | 4000 | 0.7531 | -0.9330 | -1.0793 | 0.5625 | 0.1463 | -127.4711 | -107.2016 | -16.6020 | -16.4468 |
| 0.7366 | 0.3 | 6000 | 0.7961 | -1.0751 | -1.1163 | 0.5033 | 0.0412 | -127.5638 | -107.5568 | -16.3735 | -16.2226 |
| 0.8288 | 0.4 | 8000 | 0.7860 | -1.0546 | -1.1457 | 0.5559 | 0.0911 | -127.6372 | -107.5056 | -16.4345 | -16.2588 |
| 0.7272 | 0.5 | 10000 | 0.7767 | -1.1017 | -1.2133 | 0.5658 | 0.1116 | -127.8062 | -107.6233 | -16.2325 | -16.0800 |
| 0.8404 | 0.6 | 12000 | 0.8029 | -1.2133 | -1.2948 | 0.5329 | 0.0815 | -128.0099 | -107.9024 | -15.9207 | -15.7572 |
| 0.8224 | 0.7 | 14000 | 0.7702 | -1.0578 | -1.1817 | 0.5625 | 0.1239 | -127.7272 | -107.5137 | -15.9578 | -15.7879 |
| 0.7267 | 0.8 | 16000 | 0.7929 | -1.1534 | -1.2255 | 0.5559 | 0.0721 | -127.8368 | -107.7526 | -16.0551 | -15.8740 |
| 0.6329 | 0.9 | 18000 | 0.8052 | -1.0222 | -1.0692 | 0.5428 | 0.0471 | -127.4461 | -107.4246 | -15.8842 | -15.7008 |
| 0.7247 | 1.0 | 20000 | 0.7764 | -1.0726 | -1.1588 | 0.5592 | 0.0861 | -127.6699 | -107.5507 | -15.9221 | -15.7406 |
Framework versions
- Transformers 4.36.2
- Pytorch 2.0.1+cu117
- Datasets 2.14.5
- Tokenizers 0.15.0