Qwen3.8-27B q8 KV cache quantization can hurt performance

A Reddit user reports that q8 KV cache quantization on Qwen3.8-27B can degrade model performance, though the issue appears tied to when and how often the quantize step runs, not the quantization itself.