Clinical Camel
Open 70B Clinical Language Model
Powered by Dr7.ai Medical AI Platform
Clinical Camel is an open medical large language model from researchers at the University of Toronto. It fine-tunes LLaMA-2 70B with QLoRA and a dialogue-based knowledge-encoding technique that turns dense medical articles into conversational training data.
Try Clinical Camel-Style Medical Dialogue
Ask clinical questions in natural conversation â the demo shows the dialogue-style medical reasoning Clinical Camel was trained for.
Unlock Full Clinical Camel Potential
Get unlimited clinical dialogue, longer conversations and API access for research and education workflows.
What is Clinical Camel?
Clinical Camel is an open-source clinical large language model released in 2023 by Augustin Toma and colleagues at the University of Toronto (Toma et al., "Clinical Camel: An Open Expert-Level Medical Language Model with Dialogue-Based Knowledge Encoding"). It is a LLaMA-2 70B model fine-tuned with QLoRA.
Its distinctive ingredient is Dialogue-Based Knowledge Encoding (DBKE): medical review articles are rewritten as multi-turn conversations, so the model absorbs clinical knowledge in the same conversational form users later interact with. The training mix also includes ShareGPT dialogues and MedQA-style questions.
At release the authors reported results on USMLE-style exams and standard medical QA benchmarks that were competitive with GPT-3.5 and ahead of earlier open medical models. Like all research LLMs it can hallucinate, and the authors position it for research and education rather than clinical deployment.
Latest Development
Released 2023 by researchers at the University of Toronto (Toma et al.)
Open weights (LLaMA-2 licence terms apply); 70B parameters; QLoRA fine-tuning
Key Features
Conversational medical knowledge in an open model
Core Capabilities
Multi-turn clinical dialogue and explanation
Answering USMLE-style and clinical vignette questions
Dialogue-Based Knowledge Encoding of medical review articles
Built on LLaMA-2 70B with parameter-efficient QLoRA fine-tuning
Open weights for research and educational use
Strong performance among open models at release
Suitable for patient-education and communication research
Transparent training recipe and data description
Performance Benchmarks
Scores reported in the Clinical Camel paper for the 70B model (not measured by Dr7.ai)
USMLE Sample Exam
64.3%Official USMLE sample questions
PubMedQA
77.9%Biomedical literature question answering
MedQA
60.7%USMLE-style multiple choice questions
MedMCQA
54.2%Indian medical entrance exam questions
Use Cases
Patient Communication
Prototype plain-language explanations of medical concepts (with clinician review).
Clinical Training
Simulated case discussions for trainees.
Virtual Consultation Research
Study conversational triage and history-taking with an open model.