Developers discuss Q4_0 vs Q8_0 KV cache quantization for local AI

Reddit thread explores quality trade-offs between Q4_0 and Q8_0 KV cache quantization for large context windows (50k+ tokens). Users report VRAM savings but mixed opinions on quality degradation at long contexts.