DGX Spark x 2 + DeepSeek V4 Flash Vision Exp I can now reach this speed for the time being.
KV cache has been switched to FP8. The maximum context is limited to 512K. It's about two steps away from the speed of 0731.
Chinese AI models are about seven months behind the best American ones, even though China has roughly 5% of the AI chips the US does.
The seven-month figure comes from Epoch AI, which has tracked the gap since 2023. The 5% is a rough estimate from anandnk24, who runs the
First time trying deepseek harness, what I saw in the chat animation
Stop renting a chatbot and calling it your company’s brain.
Someone open-sourced an enterprise knowledge system that remembers not only what happened, but exactly when it was true. It’s called Utopia. Most knowledge bases show the latest version of a fact. When something
You have reached the end of the archive
All of deepseek