diff options
Diffstat (limited to 'machines/kusanagi/default.nix')
| -rw-r--r-- | machines/kusanagi/default.nix | 5 |
1 files changed, 4 insertions, 1 deletions
diff --git a/machines/kusanagi/default.nix b/machines/kusanagi/default.nix index ff786f6..1c8ccb1 100644 --- a/machines/kusanagi/default.nix +++ b/machines/kusanagi/default.nix @@ -228,8 +228,11 @@ # KV-cache quantization (below) is a no-op without flash attention; ollama # falls back to f16 KV cache, inflating VRAM use and forcing CPU offload. OLLAMA_FLASH_ATTENTION = "1"; - OLLAMA_CONTEXT_LENGTH = "32768"; + OLLAMA_CONTEXT_LENGTH = "65536"; OLLAMA_KV_CACHE_TYPE = "q8_0"; + # Keep the model resident between agent turns; the default 5-minute + # keep-alive forces a multi-second reload after any pause. + OLLAMA_KEEP_ALIVE = "1h"; }; }; |