qvac/MedPsy-1.7B
MedPsy-1.7B is a 1.7 billion parameter text-only causal language model developed by Tether AI Research, built on Qwen3-1.7B. It is specifically optimized for medical and healthcare reasoning, designed for efficient deployment on edge and smartphone devices. The model demonstrates medical reasoning capabilities comparable to models 2-7x its size, achieving strong performance on closed-ended medical benchmarks and real-world clinical tasks.
Loading preview...
MedPsy-1.7B: Edge-Optimized Medical AI
MedPsy-1.7B, developed by Tether AI Research, is a 1.7 billion parameter text-only causal language model built upon the Qwen3-1.7B architecture. It is specifically engineered for efficient deployment on edge and smartphone devices, offering advanced medical reasoning capabilities in a compact footprint.
Key Capabilities & Performance
- Smartphone-Class Medical AI: At 1.7B parameters, it runs efficiently on mobile and edge devices.
- Strong Benchmark Performance: Achieves 62.62 on closed-ended medical benchmarks, outperforming MedGemma-1.5-4B (51.20) and matching Qwen3-4B Thinking (63.10).
- Clinical Task Excellence: Surpasses MedGemma-27B on real-world clinical tasks, scoring 70.33 on HealthBench and 54.33 on HealthBench Hard.
- Token Efficiency: Produces accurate medical answers with 1.7x fewer tokens than its base model, reducing latency and compute costs.
- Privacy-First: Supports fully on-device inference, ensuring patient data privacy.
Training & Methodology
The model was post-trained using a multi-stage pipeline involving supervised fine-tuning (SFT) on curated medical data and reinforcement learning (RL) with DAPO. This process adapted the Qwen3-1.7B (Thinking) backbone for specialized medical reasoning, leveraging large-scale synthetic corpora and high-value clinical QA datasets.
Intended Use Cases
MedPsy-1.7B is intended as a starting point for developers and researchers in healthcare. Appropriate uses include:
- Research into medical language understanding and reasoning.
- Building developer tools and prototypes for health-related applications.
- On-device medical information retrieval in privacy-sensitive environments.
Important Note: This model is not a substitute for professional medical judgment or clinical diagnosis and should always be used with appropriate disclaimers and safety measures.