keep gemma3:4b warm with 30m keep_alive (eliminate 6min cold-load timeouts)

dbaac8fc9faf98fd0ef3214fcaa0643d340bf515 · 2026-05-06 11:51:51 -0700 · steve

Files touched

M poll.py

Diff

commit dbaac8fc9faf98fd0ef3214fcaa0643d340bf515
Author: steve <steve@designerwallcoverings.com>
Date:   Wed May 6 11:51:51 2026 -0700

    keep gemma3:4b warm with 30m keep_alive (eliminate 6min cold-load timeouts)
---
 poll.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/poll.py b/poll.py
index 701ce28..66ea5f7 100755
--- a/poll.py
+++ b/poll.py
@@ -45,6 +45,7 @@ def call_ollama(system, user, timeout=360):
         'model': OLLAMA_MODEL,
         'prompt': system + '\n\n' + user + '\n\nOutput only the updated JSON object, no explanation, no markdown.',
         'stream': False,
+        'keep_alive': '30m',
         'options': {'temperature': 0.4, 'num_predict': 1500, 'num_ctx': 4096},
     }).encode()
     req = urllib.request.Request(

← 170bcf7 ignore runtime files (cursor, logs) · back to Videos Agentabrams Worker · snapshot: 1 file(s) changed, ~1 modified 9fd034f →