Deepseek's DSpark boosts AI speed by up to 85% amid US export controls
Deepseek's DSpark framework increases AI model response speed by 60-85% per user by using a small model to propose token candidates, batch-validated by a larger model.
Published 2sem1 sourceNotable
Lire en français
≈ 28s
The fact
This optimization reduces reliance on high-end US chips, partially bypassing export restrictions imposed by the United States.
The solution aims to strengthen China's technological autonomy in AI amid geopolitical tensions.
Click the link to read the original article: