- Max Sequence Length: 1024
- Batch Size: 2 (effective 16)
- Learning Rate: 0.00015
- LoRA Rank: 32
- LoRA Alpha: 64
- Target Modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj

Answer Structure

All answers follow a standardized 5-section format:

Overview & Pathophysiology - Mechanism and underlying processes
Clinical Presentation - Signs, symptoms, examination findings
Diagnostic Approach - Investigations and reasoning
Management Principles - Treatment approaches
Clinical Vignette - Realistic clinical scenario

Usage

python

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_name = "krishna195/medgemma-anatomy-v2.0"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

question = "What is the anatomical snuffbox and its clinical significance?"
prompt = f"<start_of_turn>user\n{question}<end_of_turn>\n<start_of_turn>model\n"

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.7)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Performance

Inference Speed: ~40-50 tokens/sec (A100), ~30-35 tokens/sec (T4)
Memory: 8-9GB (bfloat16), 3-4GB (4-bit)
Quality: Comprehensive structured answers, clinical reasoning

Improvements over v1.2

3.5x more training data (183 → 830 questions)
Higher LoRA rank (8 → 32) for better adaptation
More epochs (4 → 6) for deeper learning
Better regularization with increased weight decay
Comprehensive target modules for full model adaptation

License

Apache 2.0

Citation

bibtex

@misc{medgemma-anatomy-v2,
  title={MedGemma-4B Anatomy v2.0},
  author={Krishna195},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/krishna195/medgemma-anatomy-v2.0}
}