metadata
library_name: transformers
tags:
- trl
- dpo
- alignment-handbook
- generated_from_trainer
model-index:
- name: OpenELM-1_1B-DPO-full-max-6-reward
results: []
OpenELM-1_1B-DPO-full-max-6-reward
This model was trained from scratch on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 1.8412
- Rewards/chosen: -15.5625
- Rewards/rejected: -17.375
- Rewards/accuracies: 0.6035
- Rewards/margins: 1.7891
- Logps/rejected: -2024.0
- Logps/chosen: -1872.0
- Logits/rejected: 1.625
- Logits/chosen: -0.2451
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.5954 | 0.1047 | 100 | 0.6902 | -1.1172 | -1.2578 | 0.5742 | 0.1416 | -414.0 | -430.0 | -10.75 | -11.0625 |
| 0.5396 | 0.2094 | 200 | 0.6928 | -2.5156 | -2.8594 | 0.6387 | 0.3438 | -576.0 | -572.0 | -9.6875 | -10.25 |
| 0.5587 | 0.3141 | 300 | 0.7014 | -3.7188 | -4.1562 | 0.6094 | 0.4355 | -704.0 | -692.0 | -8.5625 | -9.4375 |
| 0.5067 | 0.4188 | 400 | 0.7869 | -3.875 | -4.375 | 0.5996 | 0.4961 | -724.0 | -708.0 | -14.0 | -14.9375 |
| 0.525 | 0.5236 | 500 | 0.8500 | -4.625 | -5.3438 | 0.6094 | 0.7109 | -824.0 | -780.0 | -6.8438 | -8.25 |
| 0.5173 | 0.6283 | 600 | 0.7292 | -6.25 | -6.7188 | 0.5723 | 0.4688 | -960.0 | -944.0 | -8.875 | -10.0 |
| 0.4944 | 0.7330 | 700 | 0.7881 | -5.1562 | -5.7812 | 0.6035 | 0.6445 | -868.0 | -832.0 | -6.4062 | -8.0 |
| 0.5113 | 0.8377 | 800 | 0.7106 | -4.7812 | -5.3438 | 0.6113 | 0.5586 | -824.0 | -796.0 | -9.4375 | -10.8125 |
| 0.4589 | 0.9424 | 900 | 0.8807 | -7.4375 | -8.1875 | 0.6094 | 0.7656 | -1112.0 | -1064.0 | -6.1875 | -8.125 |
| 0.1368 | 1.0471 | 1000 | 1.1006 | -8.375 | -9.4375 | 0.5879 | 1.0547 | -1232.0 | -1160.0 | -3.4531 | -5.375 |
| 0.138 | 1.1518 | 1100 | 1.0286 | -8.375 | -9.375 | 0.5977 | 0.9531 | -1224.0 | -1160.0 | -4.0938 | -5.9688 |
| 0.1376 | 1.2565 | 1200 | 1.0962 | -8.6875 | -9.75 | 0.6035 | 1.0312 | -1264.0 | -1192.0 | -1.2266 | -3.0781 |
| 0.1434 | 1.3613 | 1300 | 1.1220 | -9.375 | -10.5 | 0.5801 | 1.1172 | -1336.0 | -1256.0 | -3.7031 | -5.6875 |
| 0.1386 | 1.4660 | 1400 | 1.0638 | -9.4375 | -10.375 | 0.6230 | 0.9570 | -1328.0 | -1256.0 | -3.5 | -5.4688 |
| 0.1258 | 1.5707 | 1500 | 1.1923 | -10.5 | -11.75 | 0.6016 | 1.1953 | -1464.0 | -1368.0 | -2.4062 | -4.5625 |
| 0.1269 | 1.6754 | 1600 | 1.2009 | -9.4375 | -10.625 | 0.6074 | 1.1562 | -1352.0 | -1264.0 | -2.8438 | -5.2188 |
| 0.0967 | 1.7801 | 1700 | 1.1723 | -10.0 | -11.125 | 0.5996 | 1.0859 | -1400.0 | -1320.0 | -1.6328 | -3.6406 |
| 0.112 | 1.8848 | 1800 | 1.0807 | -9.75 | -10.75 | 0.5898 | 0.9805 | -1360.0 | -1296.0 | -2.5 | -4.5 |
| 0.1158 | 1.9895 | 1900 | 1.1470 | -10.875 | -12.0625 | 0.5938 | 1.2109 | -1496.0 | -1400.0 | -1.5391 | -3.5625 |
| 0.0172 | 2.0942 | 2000 | 1.6192 | -14.1875 | -15.6875 | 0.6055 | 1.5078 | -1864.0 | -1736.0 | 0.8438 | -1.1172 |
| 0.012 | 2.1990 | 2100 | 1.7070 | -14.6875 | -16.375 | 0.6016 | 1.6953 | -1928.0 | -1792.0 | 0.5117 | -1.4688 |
| 0.0145 | 2.3037 | 2200 | 1.6657 | -14.0625 | -15.625 | 0.5957 | 1.5547 | -1856.0 | -1728.0 | 0.6875 | -1.2891 |
| 0.0161 | 2.4084 | 2300 | 1.8217 | -15.5625 | -17.25 | 0.6035 | 1.7344 | -2016.0 | -1872.0 | 1.0 | -0.9141 |
| 0.0161 | 2.5131 | 2400 | 1.7852 | -15.0 | -16.625 | 0.6055 | 1.6641 | -1960.0 | -1824.0 | 1.6328 | -0.2471 |
| 0.0182 | 2.6178 | 2500 | 1.9600 | -16.25 | -18.125 | 0.5957 | 1.8125 | -2096.0 | -1952.0 | 1.7578 | -0.1089 |
| 0.0121 | 2.7225 | 2600 | 1.8076 | -15.125 | -16.875 | 0.6113 | 1.7656 | -1976.0 | -1832.0 | 1.4922 | -0.4238 |
| 0.016 | 2.8272 | 2700 | 1.8344 | -15.5 | -17.25 | 0.6055 | 1.7891 | -2016.0 | -1872.0 | 1.6016 | -0.2773 |
| 0.0144 | 2.9319 | 2800 | 1.8412 | -15.5625 | -17.375 | 0.6035 | 1.7891 | -2024.0 | -1872.0 | 1.625 | -0.2451 |
Framework versions
- Transformers 4.45.1
- Pytorch 2.3.0
- Datasets 3.0.1
- Tokenizers 0.20.0