metadata
library_name: transformers
license: apache-2.0
base_model: tsavage68/IE_M2_1000steps_1e7rate_SFT
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: IE_M2_1000steps_1e6rate_05beta_cSFTDPO
results: []
IE_M2_1000steps_1e6rate_05beta_cSFTDPO
This model is a fine-tuned version of tsavage68/IE_M2_1000steps_1e7rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.3743
- Rewards/chosen: -0.6213
- Rewards/rejected: -9.0827
- Rewards/accuracies: 0.4600
- Rewards/margins: 8.4614
- Logps/rejected: -59.1872
- Logps/chosen: -43.4481
- Logits/rejected: -2.8827
- Logits/chosen: -2.8204
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: 1e-06
- train_batch_size: 2
- eval_batch_size: 1
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 4
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 100
- training_steps: 1000
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.4505 | 0.4 | 50 | 0.3743 | -0.5495 | -8.1396 | 0.4600 | 7.5901 | -57.3010 | -43.3045 | -2.8855 | -2.8234 |
| 0.3812 | 0.8 | 100 | 0.3743 | -0.5954 | -8.5105 | 0.4600 | 7.9152 | -58.0429 | -43.3963 | -2.8832 | -2.8209 |
| 0.3119 | 1.2 | 150 | 0.3743 | -0.6132 | -8.8178 | 0.4600 | 8.2046 | -58.6574 | -43.4318 | -2.8827 | -2.8203 |
| 0.3639 | 1.6 | 200 | 0.3743 | -0.6125 | -8.8526 | 0.4600 | 8.2401 | -58.7270 | -43.4304 | -2.8828 | -2.8204 |
| 0.4332 | 2.0 | 250 | 0.3743 | -0.6117 | -8.9373 | 0.4600 | 8.3256 | -58.8965 | -43.4290 | -2.8831 | -2.8207 |
| 0.3986 | 2.4 | 300 | 0.3743 | -0.6002 | -8.9298 | 0.4600 | 8.3296 | -58.8814 | -43.4059 | -2.8831 | -2.8207 |
| 0.3986 | 2.8 | 350 | 0.3743 | -0.6140 | -8.9769 | 0.4600 | 8.3630 | -58.9757 | -43.4335 | -2.8828 | -2.8204 |
| 0.4505 | 3.2 | 400 | 0.3743 | -0.6256 | -8.9903 | 0.4600 | 8.3647 | -59.0024 | -43.4567 | -2.8829 | -2.8205 |
| 0.4505 | 3.6 | 450 | 0.3743 | -0.6066 | -8.9960 | 0.4600 | 8.3894 | -59.0138 | -43.4187 | -2.8831 | -2.8207 |
| 0.4332 | 4.0 | 500 | 0.3743 | -0.6183 | -9.0594 | 0.4600 | 8.4410 | -59.1406 | -43.4422 | -2.8830 | -2.8206 |
| 0.3292 | 4.4 | 550 | 0.3743 | -0.6163 | -9.0734 | 0.4600 | 8.4571 | -59.1686 | -43.4381 | -2.8830 | -2.8206 |
| 0.3639 | 4.8 | 600 | 0.3743 | -0.6171 | -9.0601 | 0.4600 | 8.4430 | -59.1421 | -43.4397 | -2.8828 | -2.8204 |
| 0.4505 | 5.2 | 650 | 0.3743 | -0.6207 | -9.0642 | 0.4600 | 8.4435 | -59.1503 | -43.4470 | -2.8830 | -2.8207 |
| 0.4505 | 5.6 | 700 | 0.3743 | -0.6061 | -9.0651 | 0.4600 | 8.4589 | -59.1519 | -43.4178 | -2.8831 | -2.8206 |
| 0.3639 | 6.0 | 750 | 0.3743 | -0.6217 | -9.0731 | 0.4600 | 8.4514 | -59.1681 | -43.4490 | -2.8829 | -2.8206 |
| 0.2426 | 6.4 | 800 | 0.3743 | -0.6241 | -9.0805 | 0.4600 | 8.4564 | -59.1827 | -43.4537 | -2.8829 | -2.8205 |
| 0.5025 | 6.8 | 850 | 0.3743 | -0.6248 | -9.0702 | 0.4600 | 8.4454 | -59.1623 | -43.4552 | -2.8827 | -2.8204 |
| 0.3119 | 7.2 | 900 | 0.3743 | -0.6258 | -9.0760 | 0.4600 | 8.4502 | -59.1739 | -43.4571 | -2.8826 | -2.8203 |
| 0.3466 | 7.6 | 950 | 0.3743 | -0.6208 | -9.0821 | 0.4600 | 8.4613 | -59.1860 | -43.4471 | -2.8827 | -2.8204 |
| 0.3812 | 8.0 | 1000 | 0.3743 | -0.6213 | -9.0827 | 0.4600 | 8.4614 | -59.1872 | -43.4481 | -2.8827 | -2.8204 |
Framework versions
- Transformers 4.44.2
- Pytorch 2.0.0+cu117
- Datasets 3.0.0
- Tokenizers 0.19.1