metadata
base_model: meta-llama/Meta-Llama-3-8B-Instruct
library_name: peft
license: llama3
tags:
- trl
- kto
- generated_from_trainer
model-index:
- name: kto-aligned-model-lora
results: []
kto-aligned-model-lora
This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B-Instruct on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.5001
- Eval/rewards/chosen: 0.1739
- Eval/logps/chosen: -0.4029
- Eval/rewards/rejected: 0.1787
- Eval/logps/rejected: -0.0087
- Eval/rewards/margins: -0.0048
- Eval/kl: 1.7305
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: 0.0001
- train_batch_size: 1
- eval_batch_size: 4
- seed: 42
- gradient_accumulation_steps: 12
- total_train_batch_size: 12
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 5
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | |
|---|---|---|---|---|
| 0.5 | 1.0 | 9 | 0.5000 | 1.2364 |
| 0.4994 | 2.0 | 18 | 0.5002 | 1.7169 |
| 0.4985 | 3.0 | 27 | 0.5003 | 1.7311 |
| 0.4981 | 4.0 | 36 | 0.5002 | 1.7306 |
| 0.4976 | 5.0 | 45 | 0.5001 | 1.7305 |
Framework versions
- PEFT 0.11.1
- Transformers 4.42.2
- Pytorch 2.2.0
- Datasets 2.20.0
- Tokenizers 0.19.1