metadata
library_name: transformers
license: apache-2.0
base_model: alignment-handbook/zephyr-7b-sft-full
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: zephyr-7b-dpo-full
results: []
zephyr-7b-dpo-full
This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.5006
- Rewards/chosen: -1.0395
- Rewards/rejected: -2.0108
- Rewards/accuracies: 0.7852
- Rewards/margins: 0.9713
- Logps/rejected: -463.7658
- Logps/chosen: -366.5239
- Logits/rejected: 0.6358
- Logits/chosen: -0.3905
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-07
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- gradient_accumulation_steps: 2
- total_train_batch_size: 128
- 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: 1
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.5644 | 0.2092 | 100 | 0.5693 | -0.4760 | -0.9833 | 0.75 | 0.5073 | -361.0131 | -310.1760 | -1.6463 | -1.8042 |
| 0.5482 | 0.4184 | 200 | 0.5285 | -0.5789 | -1.3324 | 0.7812 | 0.7535 | -395.9196 | -320.4612 | -1.1108 | -1.6512 |
| 0.4952 | 0.6276 | 300 | 0.5067 | -1.0198 | -1.9482 | 0.7734 | 0.9284 | -457.5016 | -364.5515 | 0.5574 | -0.3940 |
| 0.5037 | 0.8368 | 400 | 0.5006 | -1.0395 | -2.0108 | 0.7852 | 0.9713 | -463.7658 | -366.5239 | 0.6358 | -0.3905 |
Framework versions
- Transformers 4.45.1
- Pytorch 2.4.0a0+f70bd71a48.nv24.06
- Datasets 2.18.0
- Tokenizers 0.20.0