---
library_name: peft
license: other
base_model: /raid/HUB_LLM/080225_vi_test_llama33_70b_instruct/checkpoint-2568/SLERP-1.5/
tags:
- llama-factory
- lora
- generated_from_trainer
model-index:
- name: dpo
  results: []
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# dpo

This model is a fine-tuned version of [/raid/HUB_LLM/080225_vi_test_llama33_70b_instruct/checkpoint-2568/SLERP-1.5/](https://huggingface.co//raid/HUB_LLM/080225_vi_test_llama33_70b_instruct/checkpoint-2568/SLERP-1.5/) on the 60k dataset.

## 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: 9e-06
- train_batch_size: 2
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- total_train_batch_size: 16
- total_eval_batch_size: 64
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: cosine_with_min_lr
- lr_scheduler_warmup_steps: 10
- num_epochs: 3.0

### Training results



### Framework versions

- PEFT 0.12.0
- Transformers 4.47.1
- Pytorch 2.5.1+cu124
- Datasets 3.1.0
- Tokenizers 0.21.0