Google Research introduces the open-domain conversational model Meena
Google Research introduced the 2.6-billion-parameter open-domain conversational model Meena but did not release an external demo or model weights.
What happened
Event details
Google Research introduced Meena on January 28, 2020. The model used an Evolved Transformer sequence-to-sequence architecture with one encoder block and 13 decoder blocks, trained on 341 GB of public social media conversations. Google said it was still evaluating the risks and benefits and therefore did not release an external demo or model checkpoint.
Assessment
Why it matters
Meena advanced research on open-domain dialogue models and human evaluation of conversational quality, including the Sensibleness and Specificity Average metric.
Availability
Access notes
The research was public, but no external demo or model weights were released.