b7df4880c8
- LLM JSON truncation: add _repair_truncated_json() to handle max_tokens cutoff - Ontology generation: increase max_tokens 4096→8192 to prevent truncation - Config generation: parallel batch processing (3 threads × 30 agents/batch) for ~3x speedup - Rate limit: reduce semaphore 30→8 to avoid GLM API 429 storms - HuggingFace offline: set HF_HUB_OFFLINE=1 to skip unreachable hf-mirror.com - Simulation recovery: auto-reset stuck "preparing" states on Flask restart - Prepare endpoint: handle concurrent prepare requests gracefully - Frontend: handle page-refresh-during-prepare edge case - API: return 429 instead of 500 for LLM rate limit errors |
||
|---|---|---|
| .. | ||
| __init__.py | ||
| file_parser.py | ||
| llm_client.py | ||
| locale.py | ||
| logger.py | ||
| retry.py | ||
| token_tracker.py | ||
| zep_paging.py | ||