Signal
New fine-tuned models encode contemplative teacher reasoning without prompts
A developer has fine-tuned large language models Qwen3-8B and Phi-4 14B to emulate contemplative and meditation teachers' reasoning styles without system prompts or retrieval augmentation.
reddit
modelstooling
Evidence locked
Today's free sample is only available for the edition's flagship signal.
Evidence preview
- Huggingface project page for Meditation-Agent-8B-GGUF (via Reddit)huggingface.co
- Encoding contemplative teacher "movement patterns" into Qwen3-8B & Phi-4 via structured reasoning atoms (via Reddit)[Project] A-LoRA fine-tuning