DeepSeek-V2 (MoE-236B)

Training compute

10²⁴ FLOP

Parameters

236B

Published

May 7, 2024

DeepSeek-V2 (MoE-236B) is an AI model developed by DeepSeek (China), first published in May 2024. It works in the language domain, on tasks such as language modeling/generation, chat and code generation.

Training it took an estimated 10²⁴ FLOP of compute (estimation method: operation counting). The model has 236,000,000,000 parameters. It was trained on roughly 8.1T datapoints. Training ran on NVIDIA H800 SXM5.

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

Full record

Organization: DeepSeek
Country of organization: China
Domain: Language
Task: Language modeling/generation, Chat, Code generation
Training compute: 10²⁴ FLOP
Compute estimation method: Operation counting
Parameters: 236,000,000,000
Dataset size: 8.1T
Training hardware: NVIDIA H800 SXM5
Chip-hours: 172.8K
Model accessibility: Open weights (restricted use)
Open weights: Yes
Epoch confidence: Confident

Benchmark results

01

Epoch Capabilities Index

ECI Score

124.6