metadata
library_name: transformers
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: OpenELM-1_1B-DPO-full-max-8-reward
results: []
OpenELM-1_1B-DPO-full-max-8-reward
This model was trained from scratch on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 1.8180
- Rewards/chosen: -14.9375
- Rewards/rejected: -17.25
- Rewards/accuracies: 0.6270
- Rewards/margins: 2.3125
- Logps/rejected: -2016.0
- Logps/chosen: -1816.0
- Logits/rejected: 1.4375
- Logits/chosen: -0.3125
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.5108 | 0.1047 | 100 | 0.6708 | -1.2344 | -1.4531 | 0.6094 | 0.2178 | -434.0 | -442.0 | -10.375 | -10.8125 |
| 0.4452 | 0.2094 | 200 | 0.7220 | -2.8281 | -3.2812 | 0.6172 | 0.4434 | -616.0 | -600.0 | -9.625 | -10.1875 |
| 0.4197 | 0.3141 | 300 | 0.7966 | -3.9844 | -4.5938 | 0.6055 | 0.6133 | -748.0 | -716.0 | -7.5 | -8.375 |
| 0.4624 | 0.4188 | 400 | 0.7233 | -3.5625 | -4.0625 | 0.6055 | 0.4941 | -692.0 | -676.0 | -11.3125 | -12.3125 |
| 0.373 | 0.5236 | 500 | 0.8017 | -4.5 | -5.1875 | 0.6270 | 0.6758 | -808.0 | -768.0 | -7.3125 | -8.6875 |
| 0.418 | 0.6283 | 600 | 0.8381 | -4.4688 | -5.2812 | 0.6133 | 0.7812 | -816.0 | -768.0 | -9.8125 | -11.0 |
| 0.4 | 0.7330 | 700 | 0.7900 | -4.3125 | -5.125 | 0.6465 | 0.8008 | -800.0 | -752.0 | -10.75 | -11.6875 |
| 0.384 | 0.8377 | 800 | 0.8198 | -5.25 | -6.0312 | 0.6367 | 0.7773 | -892.0 | -844.0 | -7.2188 | -8.6875 |
| 0.3594 | 0.9424 | 900 | 1.0285 | -7.25 | -8.4375 | 0.6289 | 1.1719 | -1128.0 | -1040.0 | -5.7188 | -7.0938 |
| 0.0833 | 1.0471 | 1000 | 1.1501 | -9.25 | -10.375 | 0.6055 | 1.1562 | -1328.0 | -1240.0 | -2.5781 | -4.0938 |
| 0.0953 | 1.1518 | 1100 | 1.2000 | -9.5 | -10.75 | 0.5938 | 1.2422 | -1360.0 | -1264.0 | -2.0 | -3.5938 |
| 0.1198 | 1.2565 | 1200 | 1.2179 | -8.5625 | -9.75 | 0.6074 | 1.1641 | -1264.0 | -1176.0 | -2.9375 | -4.9688 |
| 0.1291 | 1.3613 | 1300 | 1.1658 | -8.75 | -10.0625 | 0.6133 | 1.2969 | -1296.0 | -1192.0 | -2.6562 | -4.5938 |
| 0.0877 | 1.4660 | 1400 | 1.1249 | -9.0 | -10.375 | 0.625 | 1.3438 | -1320.0 | -1216.0 | -1.3359 | -3.2969 |
| 0.1044 | 1.5707 | 1500 | 1.1523 | -9.75 | -11.375 | 0.6484 | 1.6328 | -1424.0 | -1296.0 | 0.3184 | -1.5312 |
| 0.0798 | 1.6754 | 1600 | 1.3625 | -11.6875 | -13.375 | 0.6172 | 1.6562 | -1624.0 | -1488.0 | -0.3613 | -2.2344 |
| 0.0847 | 1.7801 | 1700 | 1.3074 | -12.25 | -13.6875 | 0.6211 | 1.4375 | -1656.0 | -1544.0 | 0.5781 | -1.0938 |
| 0.1018 | 1.8848 | 1800 | 1.1160 | -9.4375 | -10.8125 | 0.6387 | 1.4219 | -1376.0 | -1264.0 | -1.4219 | -3.3125 |
| 0.0649 | 1.9895 | 1900 | 1.3142 | -10.75 | -12.4375 | 0.6211 | 1.6797 | -1536.0 | -1392.0 | -1.0312 | -3.0469 |
| 0.0155 | 2.0942 | 2000 | 1.7397 | -14.4375 | -16.5 | 0.6309 | 2.0781 | -1944.0 | -1768.0 | 0.6602 | -1.1641 |
| 0.0091 | 2.1990 | 2100 | 1.7227 | -14.0 | -16.125 | 0.625 | 2.1562 | -1904.0 | -1720.0 | 0.5039 | -1.3672 |
| 0.0101 | 2.3037 | 2200 | 1.8446 | -15.0625 | -17.25 | 0.6406 | 2.2344 | -2016.0 | -1824.0 | 0.9844 | -0.8359 |
| 0.0145 | 2.4084 | 2300 | 1.7911 | -14.5 | -16.75 | 0.6289 | 2.2656 | -1968.0 | -1768.0 | 1.1719 | -0.6680 |
| 0.0114 | 2.5131 | 2400 | 1.7978 | -14.75 | -17.0 | 0.6270 | 2.2812 | -1992.0 | -1792.0 | 1.2344 | -0.5430 |
| 0.012 | 2.6178 | 2500 | 1.7940 | -14.6875 | -17.0 | 0.6309 | 2.2812 | -1984.0 | -1784.0 | 1.2656 | -0.5234 |
| 0.0075 | 2.7225 | 2600 | 1.8009 | -14.75 | -17.0 | 0.6309 | 2.2656 | -1992.0 | -1792.0 | 1.3594 | -0.4062 |
| 0.0142 | 2.8272 | 2700 | 1.8148 | -14.9375 | -17.25 | 0.6309 | 2.2969 | -2016.0 | -1816.0 | 1.4453 | -0.3027 |
| 0.0102 | 2.9319 | 2800 | 1.8180 | -14.9375 | -17.25 | 0.6270 | 2.3125 | -2016.0 | -1816.0 | 1.4375 | -0.3125 |
Framework versions
- Transformers 4.45.1
- Pytorch 2.3.0
- Datasets 3.0.1
- Tokenizers 0.20.0