Model Card for Julia, an AI for Medical Reasoning
Julia Medical Reasoning is a fine-tuned version of Google's Gemma-3 model, optimized for clinical reasoning, diagnostic support, and medical question answering in English. It has been adapted through supervised fine-tuning using a curated dataset composed of medical case studies, question-answer pairs, and evidence-based medicine protocols.
Model Details
Model Description
- Developed by: Miguel Araújo Julio
- Shared by: Miguel Araújo Julio
- Model type: Causal Language Model
- Language(s) (NLP): English
- License: Apache 2.0
- Finetuned from model: google/gemma-3-1b-it
Model Sources [optional]
Uses
Direct Use
- Medical education and training.
- Assisting clinicians with reasoning through differential diagnoses.
- Generating answers to patient queries and common clinical questions.
Downstream Use
- Integration into clinical decision support tools.
- Augmenting chatbot interfaces for hospitals or telemedicine platforms.
Out-of-Scope Use
- Final medical diagnosis or treatment recommendation without human oversight.
- Use in high-risk clinical environments without regulatory clearance.
Bias, Risks, and Limitations
The model may reproduce biases found in the training data and should not be considered a replacement for licensed medical professionals. There is a risk of hallucinated or outdated information being presented as fact.
Recommendations
Users should validate outputs against trusted medical sources and consult with qualified professionals before making clinical decisions.
How to Get Started with the Model
1from transformers import AutoModelForCausalLM, AutoTokenizer
2
3model = AutoModelForCausalLM.from_pretrained("your-username/julia-medical-reasoning")
4tokenizer = AutoTokenizer.from_pretrained("your-username/julia-medical-reasoning")
5
6inputs = tokenizer("What are common symptoms of diabetes?", return_tensors="pt")
7outputs = model.generate(**inputs)
8print(tokenizer.decode(outputs[0]))
Training Details
Training Data
The model was trained using a mix of open medical question-answer datasets, synthetic case-based reasoning examples, and filtered PubMed articles.
Training Procedure
Preprocessing
Data was filtered to remove out-of-domain or unsafe content, and pre-tokenized using Gemma's tokenizer.
Training Hyperparameters
- Training regime: bf16 mixed precision
Speeds, Sizes, Times [optional]
- Fine-tuned for 3 epochs on a L4x4 GPU.
Evaluation
Testing Data, Factors & Metrics
Testing Data
Curated benchmark sets of medical reasoning and multiple-choice questions (FreedomIntelligence/medical-o1-reasoning-SFT).
Technical Specifications [optional]
Model Architecture and Objective
Decoder-only transformer architecture following Gemma specifications.
Compute Infrastructure
Hardware
L4 x 4 (96 GB)
Software
- PyTorch 2.1
- PEFT 0.14.0
- Transformers 4.40
Citation [optional]
BibTeX:
1@misc{julia2025,
2 title={Julia Medical Reasoning: Fine-tuning Gemma for Medical Understanding},
3 author={Miguel Araújo Julio},
4 year={2025},
5 url={https://huggingface.co/Miguell-J/julia-medical-reasoning}
6}
Model Card Authors [optional]
Miguel Araújo Julio
Model Card Contact
Framework versions
