- model_manager: handle unclosed <think> tags (max_tokens cutoff)
- model_manager: extract content from ALL thinking blocks, not just closed ones
- model_manager: skip reasoning prefix lines (ι¦ε /ηΆε/ε―/η¨ζ·θ¦ζ±...)
- parser.py: _strip_thinking() now extracts from thinking if result empty
- Both model output and parser input handle 100% thinking content
- Expanded skip list: 8 common Chinese reasoning prefixes
feat: working pipeline with MiniCPM5-1B GGUF on CPUβ
- Text-based parser format (more reliable than JSON with 1B model)
- Three-tier fallback: text format β simple list β regex extraction
- ModelManager with thinking-tag stripping and local GGUF support
- Full pipeline tested: parse β generate β world graph (34.6s)
- Models: MiniCPM5-1B (CPU) + MiniCPM4.1-8B (GPU-ready)
