Skip to content
A Spellwright Studios publication
Spellwright News

AI Intelligence · Google

PaLM 540B (paper model)

ClosedAnnounced, no public access

PaLM’s original April 5, 2022 paper describes a 540-billion-parameter language model. This entry pins that research target rather than an API service or a later PaLM-Coder fine-tune. Public access, model-weight release and current availability are not established by the paper.

Benchmark scores

  • HellaSwag · Reported by the model's maker, Apr 5, 202283.8%#2 of 5 results
  • HellaSwag · Reported by the model's maker, Apr 5, 202283.4%#3 of 5 results
  • HumanEval · Reported by the model's maker, Apr 5, 202276.2%#1 of 4 results
  • HumanEval · Reported by the model's maker, Apr 5, 202226.2%#4 of 4 results
  • LAMBADA · Reported by the model's maker, Apr 5, 202289.7%#1 of 4 results
  • LAMBADA · Reported by the model's maker, Apr 5, 202277.9%#3 of 4 results

Access

Historical research target described in the original PaLM paper. The paper gives a 2048-token training sequence; no hosted API limit or public checkpoint is assumed. Public/API/weights-access and retirement dates remain unqualified.