LLaMA-7B

Training compute

4×10²² FLOP

Parameters

6.7B

Published

Feb 24, 2023

LLaMA-7B 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 4×10²² FLOP of compute (estimation method: operation counting). The model has 6,700,000,000 parameters. It was trained on roughly 1T datapoints. Training ran on NVIDIA A100. The compute alone is estimated at $47K 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: 4×10²² FLOP
Compute estimation method: Operation counting
Parameters: 6,700,000,000
Dataset size: 1T
Training hardware: NVIDIA A100
Chip-hours: 82.4K
Training cost (2023 USD): $47K
Model accessibility: Open weights (non-commercial)
Open weights: Yes
Citations: 19,926
Epoch confidence: Confident

Benchmark results

01	Epoch Capabilities Index	ECI Score	96.3
02	PIQA	Score	0.8
03	BoolQ	Score	0.77
04	HellaSwag	Overall accuracy	0.76
05	LAMBADA	Score	0.73
06	TriviaQA	EM	0.71
07	WinoGrande	Accuracy	0.7
08	OpenBookQA	Accuracy	0.57