chore(halo): reduce ubatch size

2026-05-21 08:47:39 +02:00 · 2026-05-21 08:47:39 +02:00 · 1376ab0ba0
commit 1376ab0ba0
parent 6c5ce8742c
1 changed files with 1 additions and 1 deletions
--- a/systems/x86_64-linux/halo/models.ini
+++ b/systems/x86_64-linux/halo/models.ini
@ -6,7 +6,7 @@ parallel            = 1
 jinja               = true
 n-gpu-layers        = 99
 threads             = 8
-ubatch-size         = 512
+ubatch-size         = 256
 cache-type-k        = bf16
 cache-type-v        = bf16
 mmap                = false