metadata
license: apache-2.0
base_model: tsavage68/UTI_M2_1000steps_1e7rate_SFT
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: UTI2_M2_100steps_1e6rate_03beta_CSFTDPO
results: []
UTI2_M2_100steps_1e6rate_03beta_CSFTDPO
This model is a fine-tuned version of tsavage68/UTI_M2_1000steps_1e7rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.5532
- Rewards/chosen: 0.1299
- Rewards/rejected: -1.4214
- Rewards/accuracies: 0.2100
- Rewards/margins: 1.5514
- Logps/rejected: -14.1121
- Logps/chosen: -4.1094
- Logits/rejected: -2.7051
- Logits/chosen: -2.7007
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: 100
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.57 | 0.3333 | 25 | 0.5476 | 0.1996 | -1.9739 | 0.2100 | 2.1735 | -15.9537 | -3.8773 | -2.6278 | -2.6270 |
| 0.5545 | 0.6667 | 50 | 0.8328 | -0.3043 | -2.4867 | 0.1900 | 2.1824 | -17.6630 | -5.5567 | -2.6282 | -2.6278 |
| 0.5718 | 1.0 | 75 | 0.6837 | -0.0010 | -4.4519 | 0.2000 | 4.4509 | -24.2136 | -4.5457 | -2.5876 | -2.5866 |
| 1.5715 | 1.3333 | 100 | 0.5532 | 0.1299 | -1.4214 | 0.2100 | 1.5514 | -14.1121 | -4.1094 | -2.7051 | -2.7007 |
Framework versions
- Transformers 4.41.2
- Pytorch 2.0.0+cu117
- Datasets 2.19.2
- Tokenizers 0.19.1