§ tools · storyline
Serving DeepSeek-V4: why million-token context is an inference systems problem
Together AI details the inference systems work required to serve DeepSeek-V4's million-token context on NVIDIA HGX B200, covering KV compression, prefix caching, and kernel optimisation.
DeepSeek-V4 makes million-token context a serving-systems problem. Together AI explores the inference work behind V4 on NVIDIA HGX B200, including compressed KV layouts, prefix caching, kernel maturity, and endpoint profiles for long-context workloads.
§ sources1 publication · timeline below