legco_ai_assistant/backend
Woody 2aca18d30e docs: add vLLM structured output fix plan
- Diagnose: vLLM ignores OpenAI-native response_format, causing NoneType error
- Diagnose: legacy fallback prompt lacks JSON instruction → empty questions
- Plan: use vLLM-native guided_json via extra_body instead of with_structured_output
- Plan: update _SEED_DECOMPOSE with JSON format instruction
- Plan: add diagnostic logging (exc_info, method, schema preview)

wip: temporary function_calling switch for vLLM (to be replaced by guided_json)
2026-04-29 16:42:23 +08:00
..
app docs: add vLLM structured output fix plan 2026-04-29 16:42:23 +08:00
uploads chore: add .gitignore with Python, Node, env, and ChromaDB exclusions 2026-04-22 15:57:04 +08:00
.env.example docs: add LLM_ENABLE_THINKING and VLLM_ENGINE to env example and README 2026-04-28 13:32:41 +08:00
pytest.ini chore(backend): update config, env template, and pytest settings 2026-04-23 13:26:08 +08:00
requirements.txt fix: vLLM chat_template_kwargs breaks LangChain structured output 2026-04-29 16:07:44 +08:00