Qwen3.8-Flash-Next on any consumer hardware: one-click install for Windows / Linux. Strata inference engine, OpenAI/Anthropic API on localhost, optional image input.
C++
Models page: the chosen conversation context and cache format count in the memory estimate (#3832)
Niko122125abbdd90cff
diff --git a/apps/host/src/lib/native-fit.ts b/apps/host/src/lib/native-fit.tsindex b34d27849..fad8c036a 100644--- a/apps/host/src/lib/native-fit.ts+++ b/apps/host/src/lib/native-fit.ts@@ -35,7 +35,7 @@ export function memoryNeeded(v: FitInput): number {const KV_BYTES: Record<string, number> = { fp16: 2112, int8: 1056, q4_0: 576, k8v4: 816 };const KV_LAYERS = 13, KV_FACTOR = 1.054, BASE_CONTEXT = 32768;export function cacheGb(context: number, kv = 'int8'): number {- return context * KV_LAYERS * (KV_BYTES[String(kv).toLowerCase()] ?? KV_BYTES.int8) * KV_FACTOR / 2 ** 30;+ return (121 + context * KV_LAYERS * (KV_BYTES[String(kv).toLowerCase()] ?? KV_BYTES.int8) * KV_FACTOR / 2 ** 20) / 1024; // 121 MiB fixed per session}export interface CacheChoice { context: number; kv?: string }