SEA-LION V3 Llama3.1 70B

Training compute

8×10²⁴ FLOP

Parameters

70B

Published

Dec 19, 2024

SEA-LION V3 Llama3.1 70B is an AI model developed by AI Singapore (Singapore), first published in December 2024. It works in the language domain, on tasks such as question answering, chat and language modeling/generation.

Training it took an estimated 8×10²⁴ FLOP of compute (estimation method: hardware,operation counting). The model has 70,000,000,000 parameters. It was trained on roughly 200B datapoints. Training ran on 192 NVIDIA H200 SXM,NVIDIA H100 SXM5 80GB for about 136 hours.

Access: Open weights (restricted use). Its weights are openly available. It is built on top of Llama 3.1-70B. Epoch AI rates the confidence of this record as confident.

Full record

Organization: AI Singapore
Country of organization: Singapore
Domain: Language
Task: Question answering, Chat, Language modeling/generation
Training compute: 8×10²⁴ FLOP
Compute estimation method: Hardware, Operation counting
Parameters: 70,000,000,000
Dataset size: 200B
Training hardware: NVIDIA H200 SXM, NVIDIA H100 SXM5 80GB
Chips used: 192
Training time: 136 h
Numerical format: BF16
Model accessibility: Open weights (restricted use)
Open weights: Yes
Base model: Llama 3.1-70B
Epoch confidence: Confident