Search
DeepSeek-V4 Unveils Its First Multimodal Model Weights, Rivals Opus-4.8 Following Visual Integration
3 day ago /
Read about 0 minute
Author：小编
After relentlessly refining its reasoning, coding, and agent functionalities, DeepSeek has now expanded the V4 version to incorporate visual input capabilities. On the morning of August 31, DeepSeek made available the model weights of DeepSeek-V4-Flash-Vision-Exp on the Hugging Face platform, empowering developers to download and deploy the model autonomously. The official repository not only furnishes the model files but also encompasses reference inference implementations for key components, including the visual encoder and Aligner, along with segments of DFlash Attention, MoE (Mixture of Experts), Hyper-Connections, and DSpark.
-
C114 Communication Network -
Communication Home
7 X 24 Track global technological trends
