oneTaken

@TerryYiDa you could simply use difference graph.

@ghost0710 I tried many combinations, with different models and different super-parameters.And I found this image is always predicted like this. In my opinion, if this input area has very little...

My machine environment is ``` Ubuntu 14.04, python conda environment, python3.6, pytorch 0.4.1, no cuda llvm-config and clang both version 5.0 ``` Same in the error message shown above, The...

What's the problem?

Yeah, you're right. It's really a small code bug.

If you find something useful, just feel free to make a PR.

I update citation and find some unexpected useful papers. And I will increase the update frequency.

可以参考下[self-instruct](https://github.com/yizhongw/self-instruct) ，与[Chinese-LLaMA-Alpaca](https://github.com/ymcui/Chinese-LLaMA-Alpaca) 这两个repo，核心思想是一样的

一份lora 权重 + 基础的模型就是一个完整的模型。 两份lora 权重就是两个大模型，互斥的。

Oh.... I got it , thanks for your kind reply.