DeepSeek: The Emerging Chinese Start-Up Revolutionizing AI Model Training

Technology

1 year ago

News sources:

Credited from: SCMP

DeepSeek, a Chinese start-up, is gaining recognition for its innovation in the open-source large language model (LLM) sector.
The firm's latest model, DeepSeek V3, boasts 671 billion parameters and was developed with significantly lower resources compared to leading companies.
DeepSeek managed to train its model in just 2.78 million GPU hours, far less than the 30.8 million GPU hours used by competitors like Meta.
The start-up’s cost-effective approach demonstrates how resource constraints can drive technological breakthroughs in AI.
Despite some controversy regarding misidentification issues, DeepSeek V1 remains the most popular AI model on Hugging Face.

For more details, visit the original article here.