Hacker News new | ask | show | jobs
by nicman23 32 days ago
> 260k context

with a single 5090?
1 comments
kgeist 31 days ago
Yep, Gated DeltaNet in Qwen3.6 requires much less VRAM for the KV cache than previous generations. Plus the KV cache is 8-bit.
link
nicman23 31 days ago
is it in llama.cpp?
link