31 lines
747 B
Markdown
31 lines
747 B
Markdown
|
---
|
||
|
|
base_model: ogwata/exp26-sft-r16-merged
|
||
|
|
datasets:
|
||
|
|
- u-10bei/dpo-dataset-qwen-cot
|
||
|
|
language:
|
||
|
|
- en
|
||
|
|
license: apache-2.0
|
||
|
|
library_name: transformers
|
||
|
|
pipeline_tag: text-generation
|
||
|
|
tags:
|
||
|
|
- dpo
|
||
|
|
- unsloth
|
||
|
|
- qwen
|
||
|
|
- alignment
|
||
|
|
---
|
||
|
|
|
||
|
|
# exp27-dpo-r16
|
||
|
|
|
||
|
|
This model is a fine-tuned version of **ogwata/exp26-sft-r16-merged** using **Direct Preference Optimization (DPO)** via the **Unsloth** library.
|
||
|
|
|
||
|
|
This repository contains the **full-merged 16-bit weights**. No adapter loading is required.
|
||
|
|
|
||
|
|
## Training Configuration
|
||
|
|
- **Base model**: ogwata/exp26-sft-r16-merged
|
||
|
|
- **Method**: DPO (Direct Preference Optimization)
|
||
|
|
- **Epochs**: 1
|
||
|
|
- **Learning rate**: 7e-07
|
||
|
|
- **Beta**: 0.2
|
||
|
|
- **Max sequence length**: 1024
|
||
|
|
- **LoRA Config**: r=8, alpha=16 (merged into base)
|