CaLM

Training compute

2.9×10¹⁹ FLOP

Parameters

86M

Published

Dec 19, 2022

CaLM is an AI model developed by University of Oxford (United Kingdom), first published in December 2022. It works in the biology domain, on tasks such as protein or nucleotide language model (plm/nlm), protein embedding, protein property prediction and protein localization prediction.

Training it took an estimated 2.9×10¹⁹ FLOP of compute (estimation method: hardware,operation counting). The model has 86,000,000 parameters. It was trained on roughly 2.5B datapoints. Training ran on 4 NVIDIA Quadro RTX 4000 for about 960 hours.

Access: Unreleased. Its weights are not openly released. The reference paper has 34 citations. Epoch AI rates the confidence of this record as likely.

Full record

Organization: University of Oxford
Country of organization: United Kingdom
Domain: Biology
Task: Protein or nucleotide language model (pLM/nLM), Protein embedding, Protein property prediction, Protein localization prediction
Training compute: 2.9×10¹⁹ FLOP
Compute estimation method: Hardware, Operation counting
Parameters: 86,000,000
Dataset size: 2.5B
Training hardware: NVIDIA Quadro RTX 4000
Chips used: 4
Training time: 960 h
Chip-hours: 3.8K
Training power draw: 1.3 kW
Model accessibility: Unreleased
Open weights: No
Citations: 34
Epoch confidence: Likely