l294265421 / multi-turn-alpaca

Multi-turn alpaca is an extension of stanford alpaca and supports multi-turn dialogue 多轮对话版alpaca

multi-turn-alpaca

Multi-turn alpaca is an extension of stanford alpaca and supports multi-turn dialogue. Multi-turn alpaca is trained on original alpaca data AlpacaDataCleaned and multi-turn data ChatAlpaca.

Prapare Data

• process_alpaca_data_cleaned
• process_chat_alpaca
• merge_data

Training Model

• filetune
  • unzip data/training_data/training_data.txt.zip
  • nohup sh run.sh multi_turn_alpaca/training_model/finetune.py > finetune.log 2>&1 &
• inference
  • nohup sh run.sh multi_turn_alpaca/training_model/alpaca_chatbot.py > alpaca_chatbot.log 2>&1 &

Trained Weight

• Lora权重
  • 百度网盘

Cases

Chat

Code

Math

Translation

Datasets

• ChatAlpaca
• AlpacaDataCleaned
• CDial-GPT
  • 2020-NLPCC-A Large-Scale Chinese Short-Text Conversation Dataset [paper]
• dailydialog
  • 2017-IJCNLP-Dailydialog: A manually labelled multi-turn dialogue dataset
• BELLE
  • BelleGroup/generated_chat_0.4M