Models

← All models

PaLM 2-L

Google DeepMind · 2023-05-10

PaLM 2-L, also referred to by Google as PaLM 2-L-Unicorn, is the largest variant evaluated in Google's PaLM 2 technical report. It is a Transformer-based text language model trained with a tuned mixture of pre-training objectives and a more multilingual, diverse corpus than PaLM; Google reports that it is significantly smaller than PaLM-540B yet uses more training compute. Google withholds exact model-size and architecture details, and the report averages L results across its last five checkpoints rather than naming a single public checkpoint.

Benchmark scores