metadata
library_name: transformers
license: llama3.1
base_model: meta-llama/Meta-Llama-3.1-8B-Instruct
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: llama-3-8b-dpo-full
results: []
llama-3-8b-dpo-full
This model is a fine-tuned version of meta-llama/Meta-Llama-3.1-8B-Instruct on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.5582
- Rewards/chosen: -1.3603
- Rewards/rejected: -2.0529
- Rewards/accuracies: 0.7262
- Rewards/margins: 0.6926
- Logps/rejected: -600.9839
- Logps/chosen: -540.5128
- Logits/rejected: -2.7438
- Logits/chosen: -2.5853
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: 3e-07
- train_batch_size: 2
- eval_batch_size: 4
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- gradient_accumulation_steps: 8
- total_train_batch_size: 128
- total_eval_batch_size: 32
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6341 | 0.2094 | 100 | 0.6223 | -0.6021 | -0.8779 | 0.7103 | 0.2758 | -483.4830 | -464.6949 | -2.6189 | -2.4353 |
| 0.5887 | 0.4187 | 200 | 0.5796 | -1.0505 | -1.5993 | 0.7143 | 0.5488 | -555.6263 | -509.5346 | -2.6508 | -2.4854 |
| 0.5667 | 0.6281 | 300 | 0.5653 | -1.0427 | -1.6191 | 0.7222 | 0.5764 | -557.6055 | -508.7539 | -2.6684 | -2.5120 |
| 0.5803 | 0.8375 | 400 | 0.5582 | -1.3603 | -2.0529 | 0.7262 | 0.6926 | -600.9839 | -540.5128 | -2.7438 | -2.5853 |
Framework versions
- Transformers 4.45.1
- Pytorch 2.4.0+cu121
- Datasets 3.0.0
- Tokenizers 0.20.0