LLaMA-13B

Training compute

7.8×10²² FLOP

Parameters

13B

Published

Feb 27, 2023

LLaMA-13B 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 and code generation.

Training it took an estimated 7.8×10²² FLOP of compute (estimation method: operation counting). The model has 13,000,000,000 parameters. It was trained on roughly 1T datapoints. Training ran on NVIDIA A100. The compute alone is estimated at $61K in 2023 dollars.

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
Training compute: 7.8×10²² FLOP
Compute estimation method: Operation counting
Parameters: 13,000,000,000
Dataset size: 1T
Training hardware: NVIDIA A100
Training cost (2023 USD): $61K
Model accessibility: Open weights (non-commercial)
Open weights: Yes
Citations: 19,926
Epoch confidence: Confident

Benchmark results

01	Epoch Capabilities Index	ECI Score	100.2
02	PIQA	Score	0.8
03	HellaSwag	Overall accuracy	0.79
04	BoolQ	Score	0.79
05	TriviaQA	EM	0.78
06	LAMBADA	Score	0.75
07	WinoGrande	Accuracy	0.73
08	OpenBookQA	Accuracy	0.56
09	ARC AI2	Challenge score	0.53
10	MMLU	EM	0.48