model weights

[1zhangtianqing]

Can I load the weights of my trained llama model

[gaopengpjlab]

We have released all pretrained weights. If you need original llama model, please follow the guidance of llama downloading.

[1zhangtianqing]

Thank you for your answer. Can I load someone else's trained Chinese llama weights? In the past, I tried to load someone else's trained llama weights, and I also tried not to load any weights. However, during the training, there was a loss is nan. If I remember correctly, loading someone else's Chinese llama also resulted in a loss is nan. How should I load someone else's trained llama weights? Thank you again for your answer!! @gaopengpjlab

[gaopengpjlab]

Can you download chinesellama from the following link : https://github.com/OpenLMLab/OpenChineseLLaMA

[1zhangtianqing]

Okay, thank you @gaopengpjlab

[1zhangtianqing]

Hello, there's still something I don't understand, Alpaca Finetuning V1 loads a PTH format model file, and the link model file you sent is a bin format file, which can be used with Alpaca Finetuning Is v1 loading fine tuned? @gaopengpjlab

[gaopengpjlab]

We provide code to merge bin format delat with pth llama weight.

The current stable version of ImageBind-LLM is built upon Open-Chinese-LLaMA for better multilingual support. The following command downloads a pre-processed delta-version patch and automatically merges it into LLaMA weights:

python get_chinese_llama.py --llama_dir=/path/to/llama_model_weights After running, the Open-Chinese-LLaMA weights will be recovered in /path/to/llama_model_weights:

/path/to/llama_model_weights ├── 7B ├── 7B_chinese └── tokenizer.model

[adda1221]

hi, where is the get_chinese_llama.py？ I do not find in llama-adapter project.