metadata
library_name: transformers
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: OpenELM-1_1B-DPO-full-max-reward-most-similar
results: []
OpenELM-1_1B-DPO-full-max-reward-most-similar
This model was trained from scratch on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 1.4525
- Rewards/chosen: -5.5938
- Rewards/rejected: -6.125
- Rewards/accuracies: 0.5234
- Rewards/margins: 0.5078
- Logps/rejected: -900.0
- Logps/chosen: -880.0
- Logits/rejected: -16.5
- Logits/chosen: -16.75
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-05
- train_batch_size: 8
- eval_batch_size: 16
- seed: 42
- distributed_type: multi-GPU
- num_devices: 4
- gradient_accumulation_steps: 2
- total_train_batch_size: 64
- total_eval_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.0972 | 0.1047 | 100 | 0.7389 | -1.3516 | -1.5781 | 0.5527 | 0.2207 | -446.0 | -454.0 | -13.625 | -13.75 |
| 0.0661 | 0.2094 | 200 | 0.9071 | -2.4219 | -2.7969 | 0.5527 | 0.3711 | -568.0 | -560.0 | -12.6875 | -12.9375 |
| 0.0937 | 0.3141 | 300 | 0.9646 | -3.0469 | -3.375 | 0.5156 | 0.3184 | -624.0 | -624.0 | -14.6875 | -15.0 |
| 0.183 | 0.4188 | 400 | 0.9573 | -2.9688 | -3.0625 | 0.5 | 0.0898 | -596.0 | -616.0 | -13.875 | -14.125 |
| 0.0827 | 0.5236 | 500 | 1.1730 | -3.7344 | -4.125 | 0.4902 | 0.3887 | -700.0 | -692.0 | -12.5625 | -12.875 |
| 0.07 | 0.6283 | 600 | 1.0648 | -3.0 | -3.2656 | 0.5176 | 0.2695 | -616.0 | -620.0 | -19.125 | -19.0 |
| 0.0789 | 0.7330 | 700 | 0.8502 | -2.8438 | -2.9531 | 0.5078 | 0.0996 | -584.0 | -604.0 | -12.0 | -12.5 |
| 0.061 | 0.8377 | 800 | 1.2018 | -3.9531 | -4.2188 | 0.5156 | 0.2578 | -708.0 | -712.0 | -17.375 | -17.625 |
| 0.0536 | 0.9424 | 900 | 1.0159 | -3.6875 | -3.8125 | 0.5312 | 0.1299 | -672.0 | -688.0 | -18.0 | -18.125 |
| 0.0051 | 1.0471 | 1000 | 1.0946 | -4.3438 | -4.6875 | 0.5156 | 0.3379 | -756.0 | -752.0 | -15.6875 | -16.125 |
| 0.0049 | 1.1518 | 1100 | 1.2330 | -4.7188 | -5.1875 | 0.5098 | 0.4648 | -808.0 | -788.0 | -17.75 | -17.875 |
| 0.0037 | 1.2565 | 1200 | 1.2518 | -4.75 | -5.25 | 0.5156 | 0.4707 | -812.0 | -796.0 | -18.5 | -18.375 |
| 0.0122 | 1.3613 | 1300 | 1.0438 | -3.9688 | -4.3125 | 0.5312 | 0.3477 | -720.0 | -716.0 | -18.25 | -18.375 |
| 0.0082 | 1.4660 | 1400 | 1.2435 | -4.75 | -5.0625 | 0.5078 | 0.3203 | -796.0 | -792.0 | -16.875 | -17.125 |
| 0.0039 | 1.5707 | 1500 | 1.2731 | -4.875 | -5.2812 | 0.5039 | 0.4023 | -816.0 | -808.0 | -15.3125 | -15.75 |
| 0.0021 | 1.6754 | 1600 | 1.3171 | -4.875 | -5.2812 | 0.5098 | 0.4160 | -820.0 | -808.0 | -15.8125 | -16.125 |
| 0.0146 | 1.7801 | 1700 | 1.2652 | -4.625 | -5.0312 | 0.5020 | 0.4141 | -792.0 | -780.0 | -16.0 | -16.125 |
| 0.0034 | 1.8848 | 1800 | 1.2840 | -4.6875 | -4.9688 | 0.5234 | 0.3027 | -788.0 | -788.0 | -16.0 | -16.25 |
| 0.0031 | 1.9895 | 1900 | 1.2655 | -4.5312 | -4.8438 | 0.5117 | 0.3008 | -772.0 | -772.0 | -16.125 | -16.25 |
| 0.0007 | 2.0942 | 2000 | 1.3138 | -4.875 | -5.25 | 0.5078 | 0.3691 | -812.0 | -804.0 | -16.5 | -16.625 |
| 0.0209 | 2.1990 | 2100 | 1.3850 | -5.25 | -5.6562 | 0.5117 | 0.4258 | -856.0 | -844.0 | -16.75 | -16.875 |
| 0.0007 | 2.3037 | 2200 | 1.4692 | -5.5625 | -6.0625 | 0.5234 | 0.4980 | -896.0 | -876.0 | -16.625 | -16.875 |
| 0.0008 | 2.4084 | 2300 | 1.5070 | -5.8125 | -6.3438 | 0.5176 | 0.5312 | -924.0 | -900.0 | -16.5 | -16.75 |
| 0.0003 | 2.5131 | 2400 | 1.4649 | -5.625 | -6.125 | 0.5234 | 0.5039 | -900.0 | -880.0 | -16.625 | -16.75 |
| 0.0003 | 2.6178 | 2500 | 1.4368 | -5.5312 | -6.0312 | 0.5176 | 0.4980 | -892.0 | -872.0 | -16.5 | -16.75 |
| 0.0007 | 2.7225 | 2600 | 1.4452 | -5.5625 | -6.0938 | 0.5215 | 0.5039 | -896.0 | -876.0 | -16.5 | -16.75 |
| 0.0009 | 2.8272 | 2700 | 1.4519 | -5.5938 | -6.125 | 0.5234 | 0.5078 | -900.0 | -880.0 | -16.5 | -16.75 |
| 0.0005 | 2.9319 | 2800 | 1.4525 | -5.5938 | -6.125 | 0.5234 | 0.5078 | -900.0 | -880.0 | -16.5 | -16.75 |
Framework versions
- Transformers 4.44.2
- Pytorch 2.3.0
- Datasets 2.21.0
- Tokenizers 0.19.1