InnerSpace CBT Reflection LoRA
0
11 commits
1 repos using this model
1 linked in READMEs
updated Jun 8, 2026
InnerSpace CBT Reflection LoRA is a QLoRA adapter for openbmb/MiniCPM5-1B-SFT.
It is trained for the InnerSpace Hugging Face Space, a private journaling and
reflection app that turns a journal entry into a structured CBT-informed
reflection.
This adapter is not a clinical model. It does not provide diagnosis, treatment, medical advice, crisis counseling, or a replacement for licensed care.
For journal analysis, the adapter is trained to produce exactly six sections:
=== EMOTIONS ===
=== LIFE AREAS ===
=== COGNITIVE DISTORTIONS ===
=== BALANCED REFRAME ===
=== TINY NEXT STEP ===
=== REFLECTION ===
For follow-up chat, the adapter is trained to stay brief, avoid hidden reasoning tags, avoid business-style skill metrics, validate the feeling, separate feeling from evidence, and ask one grounded question.
openbmb/MiniCPM5-1B-SFTThe model should respond with supportive reflection, not certainty. It should not diagnose the user, prescribe treatment, provide crisis intervention, or claim to know whether the user's thoughts are objectively true. For immediate danger or crisis situations, users should contact local emergency services or a crisis hotline.
InnerSpace CBT Reflection LoRA
0
11 commits
1 repos using this model
1 linked in READMEs
updated Jun 8, 2026
InnerSpace CBT Reflection LoRA is a QLoRA adapter for openbmb/MiniCPM5-1B-SFT.
It is trained for the InnerSpace Hugging Face Space, a private journaling and
reflection app that turns a journal entry into a structured CBT-informed
reflection.
This adapter is not a clinical model. It does not provide diagnosis, treatment, medical advice, crisis counseling, or a replacement for licensed care.
For journal analysis, the adapter is trained to produce exactly six sections:
=== EMOTIONS ===
=== LIFE AREAS ===
=== COGNITIVE DISTORTIONS ===
=== BALANCED REFRAME ===
=== TINY NEXT STEP ===
=== REFLECTION ===
For follow-up chat, the adapter is trained to stay brief, avoid hidden reasoning tags, avoid business-style skill metrics, validate the feeling, separate feeling from evidence, and ask one grounded question.
openbmb/MiniCPM5-1B-SFTThe model should respond with supportive reflection, not certainty. It should not diagnose the user, prescribe treatment, provide crisis intervention, or claim to know whether the user's thoughts are objectively true. For immediate danger or crisis situations, users should contact local emergency services or a crisis hotline.