Running Qwen3.8-27B-Q4 at max context on a 32 GB GPU while avoiding kvcache quantization

A user successfully ran the Qwen3.8-27B-Q4 model at a maximum context of 262,144 tokens on a 32 GB GPU, while avoiding kvcache quantization. The baseline achieved 55,040 tokens with 17.13 GiB for weights and 5.65 GiB for context. Strategies like moving mmproj-to-cpu and disabling-spec significantly increased context, with the final strategy involving quantize-kv-q4 to reach the maximum context.