Whisper

Training compute

4.2×10²¹ FLOP

Parameters

1.6B

Published

Sep 21, 2022

Whisper is an AI model developed by OpenAI (United States), first published in September 2022. It works in the speech domain, on tasks such as speech recognition (asr).

Training it took an estimated 4.2×10²¹ FLOP of compute (estimation method: hardware). The model has 1,550,000,000 parameters. It was trained on roughly 12.4B datapoints.

Access: Open weights (unrestricted). Its weights are openly available. The reference paper has 7,066 citations. Epoch AI rates the confidence of this record as likely.

Full record

Organization: OpenAI
Country of organization: United States
Domain: Speech
Task: Speech recognition (ASR)
Training compute: 4.2×10²¹ FLOP
Compute estimation method: Hardware
Parameters: 1,550,000,000
Dataset size: 12.4B
Numerical format: FP16
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Citations: 7,066
Epoch confidence: Likely