metadata
library_name: transformers
tags:
- trl
- dpo
- alignment-handbook
- generated_from_trainer
model-index:
- name: OpenELM-1_1B-DPO-full-least-similar
results: []
OpenELM-1_1B-DPO-full-least-similar
This model was trained from scratch on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 1.0609
- Rewards/chosen: -3.7969
- Rewards/rejected: -4.0
- Rewards/accuracies: 0.5
- Rewards/margins: 0.2148
- Logps/rejected: -692.0
- Logps/chosen: -700.0
- Logits/rejected: -12.9375
- Logits/chosen: -13.25
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.187 | 0.1047 | 100 | 0.6760 | -0.4492 | -0.5586 | 0.5469 | 0.1084 | -344.0 | -364.0 | -14.5625 | -14.6875 |
| 0.1162 | 0.2094 | 200 | 0.6879 | -0.7734 | -0.875 | 0.5410 | 0.1016 | -376.0 | -396.0 | -11.6875 | -12.0625 |
| 0.1436 | 0.3141 | 300 | 0.7670 | -1.6562 | -1.7969 | 0.4941 | 0.1377 | -468.0 | -484.0 | -13.875 | -14.0625 |
| 0.1461 | 0.4188 | 400 | 0.7442 | -1.0469 | -1.0625 | 0.5039 | 0.0201 | -394.0 | -422.0 | -16.5 | -16.625 |
| 0.1352 | 0.5236 | 500 | 0.8131 | -1.6406 | -1.7031 | 0.5020 | 0.0630 | -460.0 | -482.0 | -15.5625 | -15.6875 |
| 0.1507 | 0.6283 | 600 | 0.8542 | -1.625 | -1.6328 | 0.4766 | 0.0096 | -452.0 | -482.0 | -17.25 | -17.375 |
| 0.1278 | 0.7330 | 700 | 0.8274 | -1.7891 | -1.9453 | 0.4980 | 0.1592 | -484.0 | -496.0 | -14.8125 | -15.0 |
| 0.1303 | 0.8377 | 800 | 0.8349 | -1.7734 | -1.7969 | 0.5195 | 0.0272 | -468.0 | -496.0 | -16.5 | -16.5 |
| 0.1614 | 0.9424 | 900 | 0.8078 | -2.2969 | -2.5 | 0.5332 | 0.1992 | -540.0 | -548.0 | -16.375 | -16.375 |
| 0.0199 | 1.0471 | 1000 | 0.8233 | -2.2656 | -2.3906 | 0.4863 | 0.1279 | -528.0 | -544.0 | -15.4375 | -15.875 |
| 0.0348 | 1.1518 | 1100 | 0.8452 | -2.0469 | -2.1562 | 0.5039 | 0.1187 | -504.0 | -524.0 | -17.0 | -17.125 |
| 0.0186 | 1.2565 | 1200 | 0.8788 | -2.9219 | -3.0312 | 0.5098 | 0.1074 | -592.0 | -612.0 | -14.75 | -15.0625 |
| 0.0277 | 1.3613 | 1300 | 0.8304 | -2.7969 | -2.8906 | 0.5137 | 0.0928 | -576.0 | -600.0 | -14.25 | -14.5 |
| 0.0212 | 1.4660 | 1400 | 0.8990 | -2.7969 | -2.9062 | 0.5 | 0.1099 | -580.0 | -600.0 | -14.25 | -14.4375 |
| 0.0333 | 1.5707 | 1500 | 0.9111 | -3.2031 | -3.2969 | 0.5215 | 0.0981 | -620.0 | -640.0 | -12.1875 | -12.625 |
| 0.0163 | 1.6754 | 1600 | 0.9215 | -3.2188 | -3.3281 | 0.4941 | 0.1104 | -620.0 | -640.0 | -11.0625 | -11.5 |
| 0.0309 | 1.7801 | 1700 | 0.9203 | -2.6719 | -2.7344 | 0.5059 | 0.0635 | -560.0 | -584.0 | -13.5625 | -13.8125 |
| 0.0228 | 1.8848 | 1800 | 0.9032 | -2.8594 | -2.9531 | 0.4941 | 0.0972 | -584.0 | -604.0 | -13.3125 | -13.5625 |
| 0.0116 | 1.9895 | 1900 | 0.9123 | -3.0156 | -3.125 | 0.5 | 0.1187 | -600.0 | -620.0 | -13.375 | -13.625 |
| 0.0011 | 2.0942 | 2000 | 0.9715 | -3.2656 | -3.4531 | 0.4980 | 0.1865 | -636.0 | -644.0 | -13.0625 | -13.3125 |
| 0.0019 | 2.1990 | 2100 | 1.0378 | -3.6719 | -3.9062 | 0.5098 | 0.2393 | -680.0 | -684.0 | -12.5 | -12.8125 |
| 0.0011 | 2.3037 | 2200 | 1.0456 | -3.7188 | -3.9375 | 0.5020 | 0.2227 | -684.0 | -692.0 | -12.8125 | -13.125 |
| 0.0009 | 2.4084 | 2300 | 1.0567 | -3.75 | -3.9688 | 0.5020 | 0.2217 | -684.0 | -692.0 | -12.9375 | -13.25 |
| 0.0022 | 2.5131 | 2400 | 1.0450 | -3.7188 | -3.9062 | 0.4961 | 0.1953 | -680.0 | -692.0 | -13.0 | -13.3125 |
| 0.0013 | 2.6178 | 2500 | 1.0499 | -3.7656 | -3.9688 | 0.5020 | 0.2080 | -684.0 | -696.0 | -12.9375 | -13.25 |
| 0.0006 | 2.7225 | 2600 | 1.0572 | -3.7812 | -3.9844 | 0.4961 | 0.2100 | -688.0 | -696.0 | -12.9375 | -13.25 |
| 0.0007 | 2.8272 | 2700 | 1.0600 | -3.7969 | -4.0 | 0.5020 | 0.2168 | -692.0 | -700.0 | -12.9375 | -13.25 |
| 0.0012 | 2.9319 | 2800 | 1.0609 | -3.7969 | -4.0 | 0.5 | 0.2148 | -692.0 | -700.0 | -12.9375 | -13.25 |
Framework versions
- Transformers 4.44.2
- Pytorch 2.3.0
- Datasets 3.0.0
- Tokenizers 0.19.1