Facebook announces open source Blender, the largest-ever open-domain chatbot
Facebook AI and machine learning department FAIR released a blog announcement that after years of research, it has built and open-sourced a new chatbot Blender. And said that this is the largest open-domain chatbot ever. “Today we’re releasing the complete model, code, and evaluation set-up, so that other AI researchers will be able to reproduce this work and continue to advance conversational AI research.”
According to reports, Blender is the first chat robot to integrate various dialogue skills (including empathy, knowledge, and personality) into one system. According to human evaluators, it is better than others in terms of engagement and it feels more human.
At the same time, Blender includes improved decoding technology, novel skill fusion, and a model with 9.4 billion parameters, which is 3.6 times higher than the largest existing system (Google chat robot Meena). In addition, it includes equally important techniques for mixing skills and detailed generation.
Facebook said that the first step in creating a chatbot is large-scale training. Blender uses 1.5 billion conversations as a training corpus. At the same time, the researchers also introduced a new Blended Skill Talk (BST) to train and evaluate the dialogue skills of the model. Blender also integrated many previous research results of Facebook.
BST includes the following skills:
Engaging use of personality (PersonaChat)
Engaging use of knowledge (Wizard of Wikipedia)
Display of empathy (Empathetic Dialogues)
Ability to blend all three seamlessly (BST)
You can get more info here.