Huysun29/cbt-gemma2-9b-v2

TEXT GENERATIONConcurrency Cost:1Model Size:9BQuant:FP8Ctx Length:16kPublished:Jun 9, 2026License:gemmaArchitecture:Transformer Cold

Huysun29/cbt-gemma2-9b-v2 is a 9 billion parameter Gemma-2-based language model fine-tuned for CBT consultation assistance. This model specializes in generating structured JSON output, including fields like risk_level, technique, and rationale, for therapeutic conversations. It is optimized for academic support in CBT, focusing on structured response generation rather than direct mental health diagnosis or treatment. The model has a context length of 16384 tokens and was trained on 26,000 CBT conversation samples.

Loading preview...

Overview

Huysun29/cbt-gemma2-9b-v2 is a 9 billion parameter language model, fine-tuned from google/gemma-2-9b-it, specifically designed to assist with Cognitive Behavioral Therapy (CBT) consultations. It excels at generating structured JSON output, which includes critical fields such as risk_level, technique, rationale, plan, response, and safety_action.

Key Capabilities

  • Structured JSON Generation: Produces well-defined JSON outputs for CBT-related conversational elements.
  • CBT Consultation Assistance: Fine-tuned on 26,000 CBT conversation samples to provide relevant and structured responses.
  • Safety Prioritization: Trained to prioritize safety and recommend professional contact or emergency services in cases of crisis or self-harm.

Training Details

The model underwent LoRA fine-tuning with parameters r=32/α=64, lr=5e-05, and a maximum sequence length of 1536. It achieved a best evaluation loss of 0.4645 at epoch 2.0.

Limitations & Safety

This model is intended as an academic support tool and should not replace mental health professionals. It is not designed for diagnosis or medication advice. Human oversight (human-in-the-loop) is crucial before deployment in real-world therapeutic settings.