LLaMA-33B

Training compute

2.7×10²³ FLOP

Parameters

32.5B

Published

Feb 27, 2023

LLaMA-33B is an AI model developed by Meta AI (United States), first published in February 2023. It works in the language domain, on tasks such as language modeling, code generation, language modeling/generation and question answering.

Training it took an estimated 2.7×10²³ FLOP of compute (estimation method: operation counting). The model has 32,500,000,000 parameters. It was trained on roughly 1.4T datapoints.

Access: Open weights (non-commercial). Its weights are openly available. The reference paper has 19,926 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Meta AI
Country of organization: United States
Domain: Language
Task: Language modeling, Code generation, Language modeling/generation, Question answering
Training compute: 2.7×10²³ FLOP
Compute estimation method: Operation counting
Parameters: 32,500,000,000
Dataset size: 1.4T
Model accessibility: Open weights (non-commercial)
Open weights: Yes
Citations: 19,926
Epoch confidence: Confident

Benchmark results

01	Epoch Capabilities Index	ECI Score	107.1
02	BoolQ	Score	0.86
03	TriviaQA	EM	0.84
04	HellaSwag	Overall accuracy	0.83
05	PIQA	Score	0.82
06	LAMBADA	Score	0.77
07	WinoGrande	Accuracy	0.76
08	ARC AI2	Challenge score	0.68
09	MMLU	EM	0.59
10	OpenBookQA	Accuracy	0.59