Phi-3.5-MoE

Training compute

3×10²³ FLOP

Parameters

60.8B

Published

Apr 23, 2024

Phi-3.5-MoE is an AI model developed by Microsoft (United States), first published in April 2024. It works in the language domain, on tasks such as language modeling/generation, translation, question answering and 2 more.

Training it took an estimated 3×10²³ FLOP of compute (estimation method: hardware,operation counting). The model has 60,800,000,000 parameters. Training ran on 512 NVIDIA H100 SXM5 80GB for about 552 hours.

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

Full record

Organization: Microsoft
Country of organization: United States
Domain: Language
Task: Language modeling/generation, Translation, Question answering, Code generation, Quantitative reasoning
Training compute: 3×10²³ FLOP
Compute estimation method: Hardware, Operation counting
Parameters: 60,800,000,000
Training hardware: NVIDIA H100 SXM5 80GB
Chips used: 512
Training time: 552 h
Training power draw: 708.4 kW
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Epoch confidence: Confident