metadata
library_name: transformers
license: llama3
base_model: tsavage68/IE_L3_1000steps_1e6rate_SFT
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: IE_L3_1000steps_1e6rate_01beta_cSFTDPO
results: []
IE_L3_1000steps_1e6rate_01beta_cSFTDPO
This model is a fine-tuned version of tsavage68/IE_L3_1000steps_1e6rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.1802
- Rewards/chosen: -0.8216
- Rewards/rejected: -13.7782
- Rewards/accuracies: 0.7400
- Rewards/margins: 12.9566
- Logps/rejected: -213.4093
- Logps/chosen: -91.0134
- Logits/rejected: -0.8670
- Logits/chosen: -0.7142
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.1913 | 0.4 | 50 | 0.1803 | -0.5046 | -8.7772 | 0.7400 | 8.2726 | -163.3993 | -87.8437 | -0.8451 | -0.7284 |
| 0.1386 | 0.8 | 100 | 0.1802 | -1.0228 | -11.9098 | 0.7400 | 10.8870 | -194.7255 | -93.0261 | -0.8546 | -0.7152 |
| 0.1386 | 1.2 | 150 | 0.1802 | -0.6732 | -12.7363 | 0.7400 | 12.0631 | -202.9905 | -89.5298 | -0.8582 | -0.7093 |
| 0.1733 | 1.6 | 200 | 0.1802 | -0.6775 | -12.8705 | 0.7400 | 12.1930 | -204.3321 | -89.5723 | -0.8611 | -0.7114 |
| 0.2253 | 2.0 | 250 | 0.1802 | -0.7149 | -13.0474 | 0.7400 | 12.3326 | -206.1017 | -89.9464 | -0.8603 | -0.7104 |
| 0.1386 | 2.4 | 300 | 0.1802 | -0.7327 | -13.0995 | 0.7400 | 12.3668 | -206.6222 | -90.1248 | -0.8593 | -0.7091 |
| 0.1213 | 2.8 | 350 | 0.1802 | -0.7598 | -13.2905 | 0.7400 | 12.5307 | -208.5327 | -90.3961 | -0.8621 | -0.7116 |
| 0.1906 | 3.2 | 400 | 0.1802 | -0.7893 | -13.4540 | 0.7400 | 12.6647 | -210.1669 | -90.6907 | -0.8653 | -0.7135 |
| 0.1906 | 3.6 | 450 | 0.1802 | -0.7880 | -13.4497 | 0.7400 | 12.6617 | -210.1245 | -90.6778 | -0.8657 | -0.7141 |
| 0.2079 | 4.0 | 500 | 0.1802 | -0.8075 | -13.6024 | 0.7400 | 12.7949 | -211.6511 | -90.8724 | -0.8653 | -0.7127 |
| 0.156 | 4.4 | 550 | 0.1802 | -0.8042 | -13.6207 | 0.7400 | 12.8165 | -211.8345 | -90.8401 | -0.8658 | -0.7138 |
| 0.1213 | 4.8 | 600 | 0.1802 | -0.8154 | -13.6478 | 0.7400 | 12.8323 | -212.1049 | -90.9520 | -0.8661 | -0.7139 |
| 0.1906 | 5.2 | 650 | 0.1802 | -0.8263 | -13.7419 | 0.7400 | 12.9156 | -213.0464 | -91.0612 | -0.8667 | -0.7144 |
| 0.2426 | 5.6 | 700 | 0.1802 | -0.8316 | -13.7569 | 0.7400 | 12.9253 | -213.1964 | -91.1135 | -0.8668 | -0.7144 |
| 0.2599 | 6.0 | 750 | 0.1802 | -0.8155 | -13.7626 | 0.7400 | 12.9471 | -213.2537 | -90.9532 | -0.8669 | -0.7141 |
| 0.1213 | 6.4 | 800 | 0.1802 | -0.8348 | -13.7975 | 0.7400 | 12.9627 | -213.6019 | -91.1453 | -0.8666 | -0.7139 |
| 0.2426 | 6.8 | 850 | 0.1802 | -0.8359 | -13.7784 | 0.7400 | 12.9425 | -213.4111 | -91.1564 | -0.8664 | -0.7143 |
| 0.1733 | 7.2 | 900 | 0.1802 | -0.8274 | -13.7943 | 0.7400 | 12.9670 | -213.5706 | -91.0716 | -0.8673 | -0.7144 |
| 0.1386 | 7.6 | 950 | 0.1802 | -0.8173 | -13.7791 | 0.7400 | 12.9618 | -213.4180 | -90.9708 | -0.8670 | -0.7140 |
| 0.156 | 8.0 | 1000 | 0.1802 | -0.8216 | -13.7782 | 0.7400 | 12.9566 | -213.4093 | -91.0134 | -0.8670 | -0.7142 |
Framework versions
- Transformers 4.44.2
- Pytorch 2.0.0+cu117
- Datasets 3.0.0
- Tokenizers 0.19.1