use 4096 kvcache.NewSWAMemCache

This commit is contained in:
jmorganca
2026-04-02 01:56:36 -07:00
parent c89280fb0c
commit f3536a356e

View File

@@ -5,7 +5,6 @@ import (
"fmt"
"image"
"log/slog"
"math"
"slices"
"time"
@@ -114,7 +113,7 @@ func New(c fs.Config) (model.Model, error) {
slidingWindowLen := int32(c.Uint("attention.sliding_window"))
m.Cache = kvcache.NewWrapperCache(
kvcache.NewSWAMemCache(slidingWindowLen, math.MaxInt32, m.Shift),
kvcache.NewSWAMemCache(slidingWindowLen, 4096, m.Shift),
kvcache.NewCausalCache(m.Shift),
)