Antigravity/notes-qpid-pro
1
1
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(ai): configure OLLAMA_KEEP_ALIVE=5m to automatically release RAM after inactivity

Browse Source
This commit is contained in:
Antigravity
2026-10-07 15:30:36 +03:00
parent 03c692fe43
commit 3b1cb85ba2
4 changed files with 8 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
app/ai.py
+4 -4
View File
@@ -251,7 +251,7 @@ async def translate_sentence_ai(
"prompt": prompt, "prompt": prompt,
"stream": False, "stream": False,
"format": "json", "format": "json",
"keep_alive": "24h", "keep_alive": config.OLLAMA_KEEP_ALIVE,
"options": { "options": {
"temperature": 0.2, "temperature": 0.2,
"num_predict": 300, "num_predict": 300,
@@ -367,10 +367,10 @@ async def warmup_ollama_model() -> bool:
client = get_ollama_client() client = get_ollama_client()
resp = await client.post( resp = await client.post(
f"{config.OLLAMA_BASE_URL}/api/generate", f"{config.OLLAMA_BASE_URL}/api/generate",
json={"model": config.OLLAMA_MODEL, "keep_alive": "24h"}, json={"model": config.OLLAMA_MODEL, "keep_alive": config.OLLAMA_KEEP_ALIVE},
) )
if resp.status_code == 200: if resp.status_code == 200:
logger.info(f"Ollama model '{config.OLLAMA_MODEL}' warmed up successfully (keep_alive=24h).") logger.info(f"Ollama model '{config.OLLAMA_MODEL}' warmed up successfully (keep_alive={config.OLLAMA_KEEP_ALIVE}).")
return True return True
else: else:
logger.warning(f"Ollama warmup returned HTTP {resp.status_code}: {resp.text}") logger.warning(f"Ollama warmup returned HTTP {resp.status_code}: {resp.text}")
@@ -567,7 +567,7 @@ async def generate_lesson_ai_debrief(
"prompt": prompt, "prompt": prompt,
"stream": False, "stream": False,
"format": "json", "format": "json",
"keep_alive": "24h", "keep_alive": config.OLLAMA_KEEP_ALIVE,
"options": { "options": {
"temperature": 0.3, "temperature": 0.3,
"num_predict": 450, "num_predict": 450,
app/config.py
+1
View File
@@ -43,4 +43,5 @@ DB_PATH.parent.mkdir(parents=True, exist_ok=True)
OLLAMA_BASE_URL = os.getenv("OLLAMA_BASE_URL", "http://127.0.0.1:11434").rstrip("/") OLLAMA_BASE_URL = os.getenv("OLLAMA_BASE_URL", "http://127.0.0.1:11434").rstrip("/")
OLLAMA_MODEL = os.getenv("OLLAMA_MODEL", "qwen2.5:3b") OLLAMA_MODEL = os.getenv("OLLAMA_MODEL", "qwen2.5:3b")
OLLAMA_KEEP_ALIVE = os.getenv("OLLAMA_KEEP_ALIVE", "5m")
docker-compose.yml
+2 -1
View File
@@ -18,6 +18,7 @@ services:
- DB_PATH=/app/data/notes.db - DB_PATH=/app/data/notes.db
- OLLAMA_BASE_URL=http://ollama:11434 - OLLAMA_BASE_URL=http://ollama:11434
- OLLAMA_MODEL=qwen2.5:3b - OLLAMA_MODEL=qwen2.5:3b
- OLLAMA_KEEP_ALIVE=5m
depends_on: depends_on:
- ollama - ollama
@@ -30,4 +31,4 @@ services:
volumes: volumes:
- ./data/ollama:/root/.ollama - ./data/ollama:/root/.ollama
environment: environment:
- OLLAMA_KEEP_ALIVE=24h - OLLAMA_KEEP_ALIVE=5m
requirements.txt
+1
View File
@@ -16,3 +16,4 @@ typing-inspection==0.4.4
typing_extensions==4.16.0 typing_extensions==4.16.0
uvicorn==0.53.0 uvicorn==0.53.0
pypdf>=5.0.0 pypdf>=5.0.0
pytest>=8.0.0