metadata
license: llama3
base_model: tsavage68/UTI_L3_1000steps_1e5rate_SFT
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: UTI2_L3_300steps_1e7rate_01beta_CSFTDPO
results: []
UTI2_L3_300steps_1e7rate_01beta_CSFTDPO
This model is a fine-tuned version of tsavage68/UTI_L3_1000steps_1e5rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.5211
- Rewards/chosen: 0.1947
- Rewards/rejected: -0.2183
- Rewards/accuracies: 0.6500
- Rewards/margins: 0.4131
- Logps/rejected: -30.6679
- Logps/chosen: -17.1558
- Logits/rejected: -1.1555
- Logits/chosen: -1.1504
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-07
- 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: 300
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6928 | 0.3333 | 25 | 0.6924 | 0.0009 | -0.0007 | 0.3600 | 0.0016 | -28.4922 | -19.0947 | -1.1524 | -1.1488 |
| 0.6893 | 0.6667 | 50 | 0.6863 | 0.0103 | -0.0035 | 0.6100 | 0.0138 | -28.5194 | -19.0000 | -1.1524 | -1.1488 |
| 0.6736 | 1.0 | 75 | 0.6701 | 0.0321 | -0.0151 | 0.6300 | 0.0471 | -28.6352 | -18.7825 | -1.1527 | -1.1490 |
| 0.622 | 1.3333 | 100 | 0.6366 | 0.0753 | -0.0439 | 0.6400 | 0.1192 | -28.9234 | -18.3503 | -1.1534 | -1.1493 |
| 0.5799 | 1.6667 | 125 | 0.5944 | 0.1218 | -0.0954 | 0.6400 | 0.2172 | -29.4390 | -17.8854 | -1.1535 | -1.1491 |
| 0.5812 | 2.0 | 150 | 0.5630 | 0.1556 | -0.1409 | 0.6500 | 0.2965 | -29.8935 | -17.5476 | -1.1544 | -1.1497 |
| 0.5284 | 2.3333 | 175 | 0.5418 | 0.1752 | -0.1786 | 0.6500 | 0.3538 | -30.2706 | -17.3511 | -1.1548 | -1.1499 |
| 0.4992 | 2.6667 | 200 | 0.5285 | 0.1875 | -0.2039 | 0.6500 | 0.3913 | -30.5232 | -17.2286 | -1.1552 | -1.1502 |
| 0.4892 | 3.0 | 225 | 0.5235 | 0.1916 | -0.2145 | 0.6500 | 0.4061 | -30.6293 | -17.1869 | -1.1554 | -1.1503 |
| 0.4895 | 3.3333 | 250 | 0.5212 | 0.1956 | -0.2171 | 0.6500 | 0.4127 | -30.6554 | -17.1470 | -1.1554 | -1.1503 |
| 0.4676 | 3.6667 | 275 | 0.5216 | 0.1945 | -0.2170 | 0.6500 | 0.4115 | -30.6547 | -17.1581 | -1.1553 | -1.1502 |
| 0.5106 | 4.0 | 300 | 0.5211 | 0.1947 | -0.2183 | 0.6500 | 0.4131 | -30.6679 | -17.1558 | -1.1555 | -1.1504 |
Framework versions
- Transformers 4.41.2
- Pytorch 2.0.0+cu117
- Datasets 2.19.2
- Tokenizers 0.19.1