IMPALA

Training compute

1.7×10²⁰ FLOP

Parameters

1.6M

Published

Feb 5, 2018

IMPALA is an AI model developed by DeepMind (United Kingdom), first published in February 2018. It works in the games domain, on tasks such as atari.

Training it took an estimated 1.7×10²⁰ FLOP of compute (estimation method: third-party estimation). The model has 1,600,000 parameters. It was trained on roughly 11.4B datapoints. Training ran on 1 NVIDIA P100 for about 100 hours. The compute alone is estimated at $53 in 2023 dollars.

Access: Unreleased. Its weights are not openly released. The reference paper has 1,829 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: DeepMind
Country of organization: United Kingdom
Domain: Games
Task: Atari
Training compute: 1.7×10²⁰ FLOP
Compute estimation method: Third-party estimation
Parameters: 1,600,000
Dataset size: 11.4B
Training hardware: NVIDIA P100
Chips used: 1
Training time: 100 h
Chip-hours: 100
Training power draw: 286 W
Training cost (2023 USD): $53
Model accessibility: Unreleased
Open weights: No
Citations: 1,829
Epoch confidence: Confident