DeepSeek Coder 33B

Training compute

4×10²³ FLOP

Parameters

33B

Published

Jan 25, 2024

DeepSeek Coder 33B is an AI model developed by DeepSeek and Peking University (China), first published in January 2024. It works in the language domain, on tasks such as code generation.

Training it took an estimated 4×10²³ FLOP of compute (estimation method: operation counting). The model has 33,000,000,000 parameters. It was trained on roughly 2T datapoints.

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

Full record

Organization: DeepSeek, Peking University
Country of organization: China
Domain: Language
Task: Code generation
Training compute: 4×10²³ FLOP
Compute estimation method: Operation counting
Parameters: 33,000,000,000
Dataset size: 2T
Model accessibility: Open weights (restricted use)
Open weights: Yes
Epoch confidence: Likely

Benchmark results

01

Epoch Capabilities Index

ECI Score

96