TD-Gammon

Training compute

1.8×10¹³ FLOP

Parameters

25K

Published

May 1, 1992

TD-Gammon is an AI model developed by IBM (United States), first published in May 1992. It works in the games domain, on tasks such as backgammon. It counts among the frontier models: the systems trained with the most compute of their moment.

Training it took an estimated 1.8×10¹³ FLOP of compute (estimation method: third-party estimation,operation counting,hardware). The model has 25,000 parameters. It was trained on roughly 6.3M datapoints.

The reference paper has 1,344 citations. Epoch AI rates the confidence of this record as speculative.

Full record

Organization: IBM
Country of organization: United States
Domain: Games
Task: Backgammon
Training compute: 1.8×10¹³ FLOP
Compute estimation method: Third-party estimation, Operation counting, Hardware
Parameters: 25,000
Dataset size: 6.3M
Citations: 1,344
Epoch confidence: Speculative