fix(diarization-ui): prevent repetition loops in Ollama generation

Adds repeat_penalty=1.15 and repeat_last_n=128 to suppress token
repetition loops (e.g. "tragen" -> "tragen" -> ...). Also caps output
via num_predict (default 4096, configurable via OLLAMA_NUM_PREDICT env
var) as a hard stop in case the model still gets stuck.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

app.py

@@ -15,6 +15,7 @@ from fastapi.responses import HTMLResponse, PlainTextResponse, Response, JSONRes
 API_BASE = os.getenv("API_BASE", "http://gx10.aquantico.lan:8093").rstrip("/")
 OLLAMA_BASE_URL = os.getenv("OLLAMA_BASE_URL", "http://gx10.aquantico.lan:11434").rstrip("/")
 OLLAMA_MODEL = os.getenv("OLLAMA_MODEL", "qwen3.5:9b")
+OLLAMA_NUM_PREDICT = int(os.getenv("OLLAMA_NUM_PREDICT", "4096"))
 DB_PATH = os.getenv("DB_PATH", "/data/ui.db")
 
 app = FastAPI(title="Diarization UI")
@@ -386,7 +387,12 @@ def _process_analysis_job(job_id: int):
 
         r = requests.post(
             f"{OLLAMA_BASE_URL}/api/generate",
-            json={"model": OLLAMA_MODEL, "prompt": llm_prompt, "stream": True, "options": {"num_ctx": num_ctx}},
+            json={"model": OLLAMA_MODEL, "prompt": llm_prompt, "stream": True, "options": {
+                "num_ctx": num_ctx,
+                "num_predict": OLLAMA_NUM_PREDICT,
+                "repeat_penalty": 1.15,
+                "repeat_last_n": 128,
+            }},
             stream=True,
             timeout=1200,
         )