Models

← All models

PaLM 2-M

Google DeepMind · 2023-05-10

PaLM 2-M is the medium variant of Google's PaLM 2 family evaluated in the May 2023 technical report. It is a Transformer-based text language model trained with a tuned mixture of pre-training objectives on a multilingual corpus spanning web documents, books, code, mathematics, conversational data, and parallel text. Google withholds the model's parameter count and detailed architecture; in the report's one-shot English QA and classification comparison, PaLM 2-M consistently outperforms PaLM 540B.

Benchmark scores