Reduce ctxcp from 32 to 16 to free ~16 GiB VRAM headroom and prevent OOM crashes during prompt processing

2026-08-18 16:11:58 -07:00
parent e5f21a5562
commit 8c391fe120
1 changed files with 1 additions and 1 deletions
@@ -43,7 +43,7 @@ fi
 : "${UBATCH_SIZE:=2048}"
 : "${CACHE_TYPE_K:=f16}"
 : "${CACHE_TYPE_V:=f16}"
-: "${CTX_CHECKPOINTS:=32}"
+: "${CTX_CHECKPOINTS:=16}"
 : "${CACHE_RAM:=8192}"
 : "${SPEC_DRAFT_N_MAX:=4}"
 : "${SPEC_DRAFT_N_MIN:=0}"