phi-3-mini 3.8B

Training compute

7.5×10²² FLOP

Parameters

3.8B

Published

Apr 23, 2024

phi-3-mini 3.8B is an AI model developed by Microsoft (United States), 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 7.5×10²² FLOP of compute (estimation method: hardware,operation counting). The model has 3,800,000,000 parameters. It was trained on roughly 3.3T datapoints. Training ran on 512 NVIDIA H100 SXM5 80GB for about 168 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: Chat, Language modeling/generation
Training compute: 7.5×10²² FLOP
Compute estimation method: Hardware, Operation counting
Parameters: 3,800,000,000
Dataset size: 3.3T
Training hardware: NVIDIA H100 SXM5 80GB
Chips used: 512
Training time: 168 h
Training power draw: 708.4 kW
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Epoch confidence: Confident

Benchmark results

01

Epoch Capabilities Index

ECI Score

117.1