marek/creator
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Training-Harness (ACO, Multi-Fidelity), Prüfstand-Benchmark, Agenten-README

Browse Source 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
team3
2026-07-04 12:47:19 +02:00
parent 8d8f6c8e51
commit 8488737303
14 changed files with 772 additions and 267 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
backend/config.py
View File

@@ -196,6 +196,7 @@ RETRY_BACKOFF = 30.0 # base seconds; backoff = base · 2^(retries-1)
MAX_RESTARTS = 2 # agent restart cap per race slot
JUDGE_CHUNK = 40 # repair: findings per judge call
EVIDENCE_PER_BLOCK = 6000 # repair: excerpt chars per fremd candidate
ABSCHLUSS_QA_LLM = 1 # 0 = Abschluss-QA ohne LLM-Judges (Training misst selbst; spart Minuten)
# Timeouts per agent step: (base seconds, seconds per block/section).
# Applies equally to all providers — whoever is too slow gets restarted or overtaken.