Huzayfah-Patel/mindbridge-phq9-hindi-merged
The Huzayfah-Patel/mindbridge-phq9-hindi-merged model is a 5.1 billion parameter Gemma 4 E2B instruction-tuned model, fine-tuned by Huzayfah Patel. It is specifically optimized for Hindi-first offline PHQ-9 and GAD-7 mental health screening, designed for community health workers in India. This model excels at generating structured JSON tool calls for scoring patient utterances in Hindi, with a context length of 32768 tokens.
Loading preview...
Overview
This model, developed by Huzayfah Patel, is a fine-tuned version of google/gemma-4-E2B-it with 5.1 billion parameters, specifically merged into fp16 weights for direct inference. It is designed for Hindi-first offline mental health screening using PHQ-9 and GAD-7 questionnaires, targeting India's ASHA community-health workers. The model is optimized for on-device deployment, including an INT8-apple quantized variant for iOS via Cactus.
Key Capabilities
- Hindi-first Screening: Specialized in processing Hindi patient utterances for mental health assessments.
- Structured Output: Emits Gemma 4 native
interpret_responsetool calls, returning a score, English rationale, and confidence. - Offline Deployment: Engineered for use in environments without internet connectivity, particularly on iOS devices.
- Robust Evaluation: Achieved significant utility lift (+25.0pp) and improved confidence calibration on a held-out evaluation set, passing predefined kill-gate thresholds.
Use Case
This model is ideal for applications requiring offline, Hindi-language mental health screening with structured output, particularly in community health settings. It serves as a component in a defense-in-depth pipeline for suicidality (Item-9) handling, with deterministic rule engines layered on top for safety. The model's design prioritizes practical, on-device deployment for healthcare professionals.