Training-Harness (ACO, Multi-Fidelity), Prüfstand-Benchmark, Agenten-README
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
@@ -1701,7 +1701,8 @@ async def _write_run_summary(topic: str, flow: Flow):
|
||||
**await db.events_run_summary(topic, run_id)}
|
||||
try: # Abschluss-QA MIT Judges: sub_dubletten/unechte werden beurteilt — erst damit
|
||||
import qa # ist note_artefakte belastbar (Kandidatenliste allein zählt nicht)
|
||||
report = await qa.qa_report(topic, llm=True)
|
||||
from config import ABSCHLUSS_QA_LLM # Training schaltet die Judges ab (misst selbst)
|
||||
report = await qa.qa_report(topic, llm=bool(ABSCHLUSS_QA_LLM))
|
||||
if report:
|
||||
summary["note"] = report["note"]
|
||||
summary["note_artefakte"] = report.get("note_artefakte")
|
||||
|
||||
Reference in New Issue
Block a user