A local medical assistant profile for Ollama, based on kwangsuklee/Qwen3.5-9B.Q4_K_M-Claude-4.6-Opus-Reasoning-Distilled-v2.

This is an Ollama Modelfile customization, not a new weight fine-tune. It keeps the original Qwen3.5 9B distilled model and adds a medical system prompt, safer generation parameters, and a /no_think template to reduce visible reasoning traces.

Run

ollama run nikitaogorodnij2006/qwen35-9b-medical

Base Model

• Base: kwangsuklee/Qwen3.5-9B.Q4_K_M-Claude-4.6-Opus-Reasoning-Distilled-v2
• Model family: Qwen3.5 9B
• Quantization: Q4_K_M
• Runtime: Ollama
• Profile type: medical instruction/profile tuning through Modelfile
• Primary language target: Russian medical study, with English prompt support

Intended Use

This profile is designed for structured medical study and clinical reasoning practice:

• pediatrics learning;
• pathogenesis, symptoms, diagnostics, and treatment principles;
• differential diagnosis practice;
• red flags and safety-oriented next steps;
• explaining medical concepts in clear language;
• handling incomplete clinical information more carefully.

It is not intended to replace a physician, provide a final diagnosis, or make emergency medical decisions.

Behavior

The model is instructed to:

• avoid inventing diagnoses, dosages, studies, percentages, and clinical criteria;
• separate facts, hypotheses, likely causes, dangerous conditions, and red flags;
• state when there is not enough information;
• account for pediatric factors such as age, body weight, development, contraindications, and safety;
• use a structured medical format when useful:
  • summary;
  • pathogenesis;
  • symptoms;
  • differential diagnosis;
  • diagnostics;
  • treatment principles;
  • red flags;
  • next steps.

Parameters

temperature: 0.25
top_p: 0.85
top_k: 40
repeat_penalty: 1.1
presence_penalty: 0.2
num_ctx: 8192

Local Comparison With Base Model

Small local comparison on 4 medical prompts:

1. pediatric respiratory distress;
2. insufficient clinical data;
3. bronchiolitis explanation;
4. false guideline / hallucination-pressure prompt.

Test environment:

• Apple M4
• 16 GB unified memory
• Ollama local runtime
• num_predict: 380

Interpretation: this profile improves instruction following, medical structure, caution with missing data, and suppression of visible reasoning traces. It does not prove higher medical accuracy in a formal clinical benchmark.

Example Prompt

Ребенок 5 лет: температура 38.8, кашель, одышка, втяжения межреберий.
Разбери по схеме: вероятно, что опасно исключить, диагностика, красные флаги, что делать сейчас.
Без дозировок.

Prompting Tips

For better answers, provide:

• age;
• body weight if treatment or dosing is discussed;
• symptom duration;
• fever pattern;
• important positive and negative symptoms;
• current medications;
• comorbidities;
• exam findings if available;
• the exact task: explanation, differential diagnosis, red flags, or study notes.

Limitations

• The model can still make mistakes.
• It can overgeneralize when the prompt lacks clinical details.
• It should not be used as the only basis for diagnosis or treatment.
• Current guidelines and medication information should be verified from authoritative medical sources.