File size: 457 Bytes
9c2a9f7 04924fe 9c2a9f7 d5a7d38 9c2a9f7 bcfcb98 3172852 331bb6b |
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 a 4 bit DPO fine-tuned MoE model for [TomGrc/FusionNet_34Bx2_MoE_v0.1](https://huggingface.co/TomGrc/FusionNet_34Bx2_MoE_v0.1)
```
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
|