File size: 497 Bytes
b86dbe5 8f41adc b86dbe5 f147175 b86dbe5 e8e558b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 |
---
tags:
- yi
- moe
license: apache-2.0
---
this is another DPO all-linear-parameter-fine-tuned MoE model for [TomGrc/FusionNet_34Bx2_MoE_v0.1](https://huggingface.co/TomGrc/FusionNet_34Bx2_MoE_v0.1)
it's trained on a H100 for one hour
```
DPO Trainer
TRL supports the DPO Trainer for training language models from preference data, as described in the paper Direct Preference Optimization: Your Language Model is Secretly a Reward Model by Rafailov et al., 2023.
```
Metrics not test!
|