fix: append /no_thinking to all Qwen prompts to disable thinking mode via prompt ef2a762 tuan3335 commited on Jun 26
fix: explicitly disable thinking mode in Qwen InferenceClient calls for both answer and wiki query optimization 8cf4282 tuan3335 commited on Jun 26
feat: use LangChain StrOutputParser for clean answer output; optimize Wikipedia search query with AI c282f35 tuan3335 commited on Jun 26
feat: improve tool selection prompt (English), log wiki and audio transcript results for debugging 6bef95f tuan3335 commited on Jun 26
refactor: switch to huggingface_hub InferenceClient for Qwen, remove local transformers usage c5208b6 tuan3335 commited on Jun 26
refactor: dùng transformers gốc cho Qwen, bổ sung accelerate vào requirements.txt 8103d43 tuan3335 commited on Jun 26
Add structured output with Pydantic, fix tool selection logic, add YouTube cookies support, disable thinking mode a9b5cb5 tuan3335 commited on Jun 26
Restore HuggingFace imports and create hybrid AI brain with HF primary + Groq fallback 1cf80b8 tuan3335 commited on Jun 26
Replace HuggingFace model with Groq free API to fix payment required error ed71eea tuan3335 commited on Jun 26
Sử dụng XML tag cho prompt AI decision, tăng độ ổn định khi parse action từ model trả về be2373e tuan3335 commited on Jun 26
