As far as i know the weight is of 160gb + 9.6gb needed for max 1 million token window + 5 gigs overhead = 175gb vram. But vllm and othere sources said "To use the full 1M context, you need 4x A100 80G" --> thats a 320gb vram ?? Am i missing something?? Sources: https://lushbinary.com/blog/deepseek-v
