metadata
library_name: transformers
tags:
- trl
- dpo
- alignment-handbook
- generated_from_trainer
model-index:
- name: OpenELM-1_1B-DPO-full-max-14-reward
results: []
OpenELM-1_1B-DPO-full-max-14-reward
This model was trained from scratch on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 1.1518
- Rewards/chosen: -4.0625
- Rewards/rejected: -4.5312
- Rewards/accuracies: 0.4980
- Rewards/margins: 0.4824
- Logps/rejected: -744.0
- Logps/chosen: -724.0
- Logits/rejected: -15.25
- Logits/chosen: -15.5625
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.05 | 0.1047 | 100 | 0.8594 | -2.0469 | -2.3906 | 0.5488 | 0.3496 | -528.0 | -524.0 | -9.3125 | -9.5625 |
| 0.0695 | 0.2094 | 200 | 0.7369 | -1.2031 | -1.3984 | 0.4883 | 0.1953 | -428.0 | -438.0 | -12.9375 | -13.3125 |
| 0.0402 | 0.3141 | 300 | 1.4284 | -4.5938 | -5.0 | 0.5195 | 0.4297 | -792.0 | -776.0 | -9.125 | -9.75 |
| 0.0373 | 0.4188 | 400 | 0.8732 | -2.3906 | -2.5156 | 0.4902 | 0.1279 | -540.0 | -556.0 | -13.375 | -13.5625 |
| 0.0289 | 0.5236 | 500 | 0.9761 | -3.5938 | -3.8906 | 0.4902 | 0.2871 | -676.0 | -680.0 | -15.0 | -15.25 |
| 0.0549 | 0.6283 | 600 | 0.9004 | -2.3594 | -2.6094 | 0.4805 | 0.2539 | -548.0 | -556.0 | -15.0625 | -15.1875 |
| 0.0385 | 0.7330 | 700 | 0.9997 | -3.125 | -3.25 | 0.4746 | 0.1299 | -612.0 | -632.0 | -11.5 | -11.875 |
| 0.0303 | 0.8377 | 800 | 1.0037 | -2.7656 | -2.9375 | 0.4785 | 0.1748 | -584.0 | -596.0 | -13.75 | -14.0625 |
| 0.0147 | 0.9424 | 900 | 1.1243 | -3.6094 | -3.7656 | 0.4824 | 0.1553 | -664.0 | -680.0 | -13.8125 | -14.1875 |
| 0.0038 | 1.0471 | 1000 | 1.0635 | -3.5781 | -3.7969 | 0.4941 | 0.2158 | -668.0 | -676.0 | -13.8125 | -14.125 |
| 0.0192 | 1.1518 | 1100 | 1.2317 | -4.1875 | -4.5625 | 0.4941 | 0.3711 | -744.0 | -736.0 | -14.0 | -14.25 |
| 0.0035 | 1.2565 | 1200 | 1.1275 | -3.8125 | -4.0938 | 0.5195 | 0.2676 | -696.0 | -700.0 | -13.6875 | -14.0625 |
| 0.0014 | 1.3613 | 1300 | 1.1072 | -3.8281 | -4.1875 | 0.5039 | 0.3672 | -708.0 | -700.0 | -10.3125 | -11.0 |
| 0.0009 | 1.4660 | 1400 | 1.2158 | -4.1562 | -4.5938 | 0.5039 | 0.4570 | -748.0 | -732.0 | -15.6875 | -16.0 |
| 0.0047 | 1.5707 | 1500 | 0.9804 | -3.4062 | -3.7656 | 0.5 | 0.3672 | -664.0 | -660.0 | -14.625 | -15.0625 |
| 0.0009 | 1.6754 | 1600 | 1.0340 | -4.0312 | -4.4688 | 0.5137 | 0.4219 | -736.0 | -724.0 | -10.6875 | -11.4375 |
| 0.0053 | 1.7801 | 1700 | 0.9808 | -3.4531 | -3.8125 | 0.5215 | 0.3730 | -672.0 | -664.0 | -16.125 | -16.25 |
| 0.0006 | 1.8848 | 1800 | 0.9781 | -3.2812 | -3.5312 | 0.5098 | 0.2578 | -640.0 | -644.0 | -16.125 | -16.25 |
| 0.0086 | 1.9895 | 1900 | 1.1759 | -4.1562 | -4.6562 | 0.5020 | 0.5195 | -756.0 | -732.0 | -15.4375 | -15.6875 |
| 0.0001 | 2.0942 | 2000 | 1.1181 | -3.8594 | -4.3125 | 0.5 | 0.4473 | -720.0 | -704.0 | -15.4375 | -15.6875 |
| 0.0145 | 2.1990 | 2100 | 1.1573 | -4.0312 | -4.5312 | 0.4980 | 0.4941 | -740.0 | -720.0 | -15.625 | -15.875 |
| 0.0002 | 2.3037 | 2200 | 1.1923 | -4.2188 | -4.7188 | 0.4961 | 0.5234 | -760.0 | -740.0 | -15.0625 | -15.4375 |
| 0.0005 | 2.4084 | 2300 | 1.1497 | -4.0 | -4.5 | 0.4902 | 0.4824 | -736.0 | -720.0 | -15.3125 | -15.5625 |
| 0.0002 | 2.5131 | 2400 | 1.1575 | -4.0312 | -4.5312 | 0.4961 | 0.4902 | -740.0 | -720.0 | -15.375 | -15.6875 |
| 0.0001 | 2.6178 | 2500 | 1.1676 | -4.0938 | -4.5938 | 0.4922 | 0.5039 | -748.0 | -728.0 | -15.25 | -15.5625 |
| 0.0014 | 2.7225 | 2600 | 1.1490 | -4.0312 | -4.5 | 0.5020 | 0.4785 | -740.0 | -720.0 | -15.3125 | -15.625 |
| 0.0002 | 2.8272 | 2700 | 1.1505 | -4.0312 | -4.5312 | 0.4961 | 0.4824 | -740.0 | -720.0 | -15.25 | -15.5625 |
| 0.0002 | 2.9319 | 2800 | 1.1518 | -4.0625 | -4.5312 | 0.4980 | 0.4824 | -744.0 | -724.0 | -15.25 | -15.5625 |
Framework versions
- Transformers 4.45.1
- Pytorch 2.3.0
- Datasets 3.0.1
- Tokenizers 0.20.0