how much difference does kv cache quants even make?
I currently run qwen flash next 3.8 iq3xxs via strata (like almost everyone lol) and I currently have kv cache at q8 with 262k ctx. Ive never actually tried any other quant and dont know how big the difference even is. Id like to have a bit more context (like 300-400k) but it doesnt fit my ram then. how bad is q4 cache? Thanks for your help Edit: Thanks for all of your suggestions, opinions and information, i will try out 8k4v as many of you recommended, otherwise ill stay at 8 for both
参与讨论