FRED-T5-XL

Training compute

2.5×10²² FLOP

Parameters

1.7B

Published

Apr 18, 2024

FRED-T5-XL is an AI model developed by Sber (Russia), first published in April 2024. It works in the language domain, on tasks such as chat and language modeling/generation.

Training it took an estimated 2.5×10²² FLOP of compute (estimation method: operation counting,hardware). The model has 1,740,000,000 parameters. It was trained on roughly 80B datapoints. Training ran on 160 NVIDIA A100,NVIDIA Tesla V100 SXM2 for about 1.9K hours.

Access: Open weights (unrestricted). Its weights are openly available. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Sber
Country of organization: Russia
Domain: Language
Task: Chat, Language modeling/generation
Training compute: 2.5×10²² FLOP
Compute estimation method: Operation counting, Hardware
Parameters: 1,740,000,000
Dataset size: 80B
Training hardware: NVIDIA A100, NVIDIA Tesla V100 SXM2
Chips used: 160
Training time: 1,920 h
Chip-hours: 265K
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Epoch confidence: Confident