sadnessdevil

Shows that the KV cache of qwen4exp-style models can sit in system RAM, so a quant that barely fits in VRAM can still use the full context length.