olmoe_densebackward / README.md
autoprogrammer's picture
Upload folder using huggingface_hub
94724ad verified
# DenseBackwardOLMoE
自定义的OLMoE模型,使用DenseBackwardOlmoeSparseMoeBlock替换原版的MoE模块,实现dense backward功能。
## 用法
```python
from transformers import AutoConfig, AutoModelForCausalLM
# 使用trust_remote_code=True加载模型
config = AutoConfig.from_pretrained("autoprogrammer/olmoe_densebackward", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("autoprogrammer/olmoe_densebackward", config=config, trust_remote_code=True)
```