|
--- |
|
library_name: transformers |
|
tags: |
|
- trl |
|
- dpo |
|
- generated_from_trainer |
|
model-index: |
|
- name: OpenELM-1_1B-DPO-full-max-4-reward |
|
results: [] |
|
--- |
|
|
|
<!-- This model card has been generated automatically according to the information the Trainer had access to. You |
|
should probably proofread and complete it, then remove this comment. --> |
|
|
|
# OpenELM-1_1B-DPO-full-max-4-reward |
|
|
|
This model was trained from scratch on an unknown dataset. |
|
It achieves the following results on the evaluation set: |
|
- Loss: 1.5952 |
|
- Rewards/chosen: -13.125 |
|
- Rewards/rejected: -14.4375 |
|
- Rewards/accuracies: 0.6035 |
|
- Rewards/margins: 1.3047 |
|
- Logps/rejected: -1728.0 |
|
- Logps/chosen: -1632.0 |
|
- Logits/rejected: 2.4062 |
|
- Logits/chosen: 0.5391 |
|
|
|
## 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 | |
|
|:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:| |
|
| 0.6374 | 0.0838 | 80 | 0.6876 | -0.6875 | -0.7773 | 0.5664 | 0.0908 | -366.0 | -386.0 | -9.8125 | -10.125 | |
|
| 0.6205 | 0.1675 | 160 | 0.6953 | -1.2266 | -1.375 | 0.5840 | 0.1475 | -426.0 | -440.0 | -11.4375 | -11.75 | |
|
| 0.6209 | 0.2513 | 240 | 0.7235 | -1.6016 | -1.7578 | 0.5762 | 0.1553 | -464.0 | -478.0 | -12.125 | -12.375 | |
|
| 0.6605 | 0.3351 | 320 | 0.7567 | -2.7969 | -3.0469 | 0.5996 | 0.2461 | -592.0 | -600.0 | -12.375 | -12.6875 | |
|
| 0.6471 | 0.4188 | 400 | 0.7148 | -2.7031 | -2.875 | 0.5801 | 0.1816 | -576.0 | -588.0 | -12.0625 | -12.5 | |
|
| 0.6121 | 0.5026 | 480 | 0.7704 | -4.1562 | -4.5625 | 0.5879 | 0.3945 | -744.0 | -736.0 | -7.4062 | -8.1875 | |
|
| 0.6059 | 0.5864 | 560 | 0.7471 | -4.125 | -4.4688 | 0.5957 | 0.3320 | -736.0 | -732.0 | -11.25 | -11.9375 | |
|
| 0.5698 | 0.6702 | 640 | 0.7281 | -3.3125 | -3.7344 | 0.6367 | 0.4297 | -660.0 | -648.0 | -14.1875 | -15.1875 | |
|
| 0.6151 | 0.7539 | 720 | 0.7413 | -2.8438 | -3.1562 | 0.5840 | 0.3105 | -604.0 | -604.0 | -12.625 | -13.0625 | |
|
| 0.5859 | 0.8377 | 800 | 0.7781 | -4.7812 | -5.2812 | 0.6074 | 0.4844 | -816.0 | -796.0 | -8.5 | -9.8125 | |
|
| 0.6049 | 0.9215 | 880 | 0.7388 | -2.8281 | -3.125 | 0.5977 | 0.2930 | -600.0 | -600.0 | -12.3125 | -12.6875 | |
|
| 0.4217 | 1.0052 | 960 | 0.7678 | -5.1875 | -5.8438 | 0.6465 | 0.6562 | -872.0 | -836.0 | -6.3125 | -7.8125 | |
|
| 0.1894 | 1.0890 | 1040 | 1.0973 | -6.9688 | -7.625 | 0.6074 | 0.6719 | -1056.0 | -1016.0 | -4.3438 | -6.0 | |
|
| 0.1959 | 1.1728 | 1120 | 0.9770 | -6.75 | -7.5 | 0.6133 | 0.7422 | -1040.0 | -992.0 | -3.9062 | -5.625 | |
|
| 0.17 | 1.2565 | 1200 | 1.0293 | -7.2188 | -7.9062 | 0.6094 | 0.6719 | -1080.0 | -1040.0 | -4.5625 | -6.0938 | |
|
| 0.1857 | 1.3403 | 1280 | 0.9556 | -7.0625 | -7.8125 | 0.5996 | 0.7578 | -1072.0 | -1024.0 | -5.75 | -7.3125 | |
|
| 0.1872 | 1.4241 | 1360 | 0.9190 | -7.25 | -8.0625 | 0.5938 | 0.8359 | -1096.0 | -1040.0 | -4.7812 | -6.4375 | |
|
| 0.1445 | 1.5079 | 1440 | 1.0569 | -9.0 | -9.9375 | 0.5996 | 0.9258 | -1280.0 | -1216.0 | -3.7656 | -5.3438 | |
|
| 0.136 | 1.5916 | 1520 | 1.0663 | -9.5625 | -10.4375 | 0.6191 | 0.9219 | -1336.0 | -1272.0 | -2.3125 | -4.0312 | |
|
| 0.1765 | 1.6754 | 1600 | 1.0288 | -8.0625 | -8.9375 | 0.6133 | 0.875 | -1184.0 | -1128.0 | -2.6562 | -4.5312 | |
|
| 0.1661 | 1.7592 | 1680 | 1.0917 | -8.0625 | -8.9375 | 0.6035 | 0.8633 | -1184.0 | -1128.0 | -2.7656 | -4.6562 | |
|
| 0.1451 | 1.8429 | 1760 | 1.0870 | -8.375 | -9.25 | 0.5957 | 0.8867 | -1216.0 | -1152.0 | -2.9688 | -4.6875 | |
|
| 0.1712 | 1.9267 | 1840 | 1.0650 | -8.6875 | -9.6875 | 0.6172 | 0.9922 | -1256.0 | -1184.0 | -2.4375 | -4.3125 | |
|
| 0.0278 | 2.0105 | 1920 | 1.0530 | -8.875 | -9.875 | 0.6152 | 0.9805 | -1272.0 | -1208.0 | -1.8906 | -3.8125 | |
|
| 0.0225 | 2.0942 | 2000 | 1.4602 | -11.5 | -12.5625 | 0.6035 | 1.0312 | -1544.0 | -1472.0 | 0.4297 | -1.5547 | |
|
| 0.0182 | 2.1780 | 2080 | 1.5544 | -12.6875 | -13.8125 | 0.5977 | 1.1172 | -1672.0 | -1592.0 | 1.7266 | -0.1621 | |
|
| 0.0385 | 2.2618 | 2160 | 1.5476 | -13.0 | -14.1875 | 0.6016 | 1.1953 | -1712.0 | -1616.0 | 1.5859 | -0.2598 | |
|
| 0.0162 | 2.3455 | 2240 | 1.5637 | -12.8125 | -13.9375 | 0.6016 | 1.1641 | -1688.0 | -1600.0 | 1.8984 | 0.0913 | |
|
| 0.0239 | 2.4293 | 2320 | 1.4822 | -11.9375 | -13.0625 | 0.5938 | 1.1797 | -1600.0 | -1512.0 | 1.2422 | -0.6172 | |
|
| 0.0264 | 2.5131 | 2400 | 1.6307 | -13.375 | -14.6875 | 0.6035 | 1.3047 | -1760.0 | -1656.0 | 2.4688 | 0.6406 | |
|
| 0.024 | 2.5969 | 2480 | 1.5421 | -12.3125 | -13.5625 | 0.5996 | 1.2188 | -1640.0 | -1552.0 | 1.7422 | -0.1631 | |
|
| 0.0217 | 2.6806 | 2560 | 1.6061 | -13.0 | -14.25 | 0.6035 | 1.2656 | -1720.0 | -1616.0 | 2.2656 | 0.3770 | |
|
| 0.0208 | 2.7644 | 2640 | 1.5995 | -13.1875 | -14.4375 | 0.6055 | 1.2812 | -1736.0 | -1632.0 | 2.5156 | 0.6602 | |
|
| 0.0206 | 2.8482 | 2720 | 1.5964 | -13.1875 | -14.4375 | 0.6055 | 1.2969 | -1736.0 | -1632.0 | 2.4688 | 0.6133 | |
|
| 0.0163 | 2.9319 | 2800 | 1.5952 | -13.125 | -14.4375 | 0.6035 | 1.3047 | -1728.0 | -1632.0 | 2.4062 | 0.5391 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.45.1 |
|
- Pytorch 2.3.0 |
|
- Datasets 3.0.1 |
|
- Tokenizers 0.20.0 |
|
|
