Compumacy/Compumacy-Shrink-24B

VISIONConcurrent Unit Cost:2Model Size:24BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 22, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Compumacy-Shrink-24B is a 24-billion parameter language model developed by Compumacy and fine-tuned by Daemontatox. It is highly specialized for clinical psychology and psychiatry, engineered to process complex clinical vignettes and generate structured, evidence-based responses. The model excels at differential diagnosis, risk assessment, and proposing treatment plans aligned with DSM-5-TR, ICD-11, and established guidelines. Its primary use is as a research tool to augment mental health professionals in clinical assessment and treatment planning.

Loading preview...

Compumacy-Shrink-24B: Specialized for Clinical Psychiatry

Compumacy-Shrink-24B is a 24-billion parameter language model developed by Compumacy and fine-tuned by Daemontatox. It is specifically designed to assist mental health professionals by processing clinical vignettes and generating structured, evidence-based assessments. The model's training utilized the Daemontatox/Psy-Data-books dataset, a comprehensive corpus of professional psychiatric literature, textbooks, and clinical guidelines, enabling it to emulate diagnostic reasoning processes.

Key Capabilities

  • Chief Complaint Analysis: Objectively summarizes presenting symptoms.
  • Differential Diagnosis: Systematically evaluates potential diagnoses against DSM-5-TR criteria.
  • Comprehensive Risk Assessment: Identifies risks related to suicide, homicide, psychosis, and substance use.
  • Evidence-Based Recommendations: Proposes treatment plans (pharmacotherapy, psychotherapy) based on guidelines from APA, WFSBP, and NICE.
  • Monitoring and Referrals: Outlines necessary follow-ups and indications for specialized care.

Intended Use and Limitations

This model is intended as a research tool to augment, not replace, the expertise of licensed clinicians. It systematically applies diagnostic criteria and references peer-reviewed literature. Users must employ a specific Alpaca-style prompt format for optimal performance. It is crucial to note that this model is not a substitute for a qualified medical professional, does not establish a therapeutic relationship, and all outputs require independent verification. Users should be aware of potential data biases, the risk of hallucinations, and must not input Protected Health Information (PHI).