This is the official repository of the paper: SPA: Towards A Computational Friendly Cloud-Base and On-Devices Collaboration Seq2seq Personalized Generation.

• Prepare well for your dataset, adjust it to the form:

{"instruction":... ,"input": ... , "output": ...}

• clone the repo

git clone [email protected]:OceannTwT/ra-isf.git

• replace the code on LlamaForCausalLM with model/modeling_SPA.py

• add your dataset on SPA/data/dataset_info.json

• tune and get the SPA model!

bash train.sh

• inference the model by running llama_SPA_predict.py, remember change the dir on your additional parameters.

• This repository is served for on-devices personalized LLMs, it could contribute to fast and reliable on-devices LLMs.

• We thank for the contributions of all co-authors and siri-china teams in this work.