DuoNeural/MedGemma-4B-IT-GGUF

MedGemma-4B-IT — GGUF Quants

Quantized GGUF versions of google/medgemma-4b-it — Google's medically specialized 4B instruction-tuned model built on the Gemma architecture. MedGemma is fine-tuned on medical literature, clinical notes, and healthcare datasets, making it a strong base for biomedical NLP, clinical decision support, and medical question answering.

Note: MedGemma is intended for research and development. It is not a substitute for professional medical advice, diagnosis, or treatment. Always consult a qualified healthcare professional.

Available Files

File	Quant	Size	Use Case
MedGemma-4B-IT-Q8_0.gguf	Q8_0	~4.8GB	Maximum quality
MedGemma-4B-IT-Q6_K.gguf	Q6_K	~3.7GB	Near-lossless
MedGemma-4B-IT-Q5_K_M.gguf	Q5_K_M	~3.2GB	High quality
MedGemma-4B-IT-Q4_K_M.gguf	Q4_K_M	~2.8GB	Recommended default
MedGemma-4B-IT-Q3_K_M.gguf	Q3_K_M	~2.3GB	Low VRAM
MedGemma-4B-IT-IQ4_XS.gguf	IQ4_XS	~2.5GB	Imatrix 4-bit
MedGemma-4B-IT-IQ3_XXS.gguf	IQ3_XXS	~1.9GB	Imatrix 3-bit
MedGemma-4B-IT-IQ2_M.gguf	IQ2_M	~1.7GB	Imatrix 2-bit
MedGemma-4B-IT-IQ1_S.gguf	IQ1_S	~1.3GB	Extreme compression
MedGemma-4B-IT-fp16.gguf	FP16	~8.5GB	Full precision
imatrix.dat	—	—	Importance matrix

Usage

# llama.cpp
./llama-cli -m MedGemma-4B-IT-Q4_K_M.gguf \
  --ctx-size 8192 -n 512 \
  -p "<start_of_turn>user\nWhat are the symptoms of type 2 diabetes?<end_of_turn>\n<start_of_turn>model\n"

# Ollama
ollama run hf.co/DuoNeural/MedGemma-4B-IT-GGUF:Q4_K_M

About MedGemma-4B

• Parameters: 4B
• Architecture: Gemma (decoder-only)
• Specialization: Medical literature, clinical NLP, biomedical QA
• License: Gemma License
• Use cases: Medical research tools, clinical decision support prototypes, biomedical text analysis

Local deployment enables privacy-preserving medical NLP without sending patient data to external APIs.

---

Quantized by DuoNeural using llama.cpp on RTX 5090.

---

DuoNeural

DuoNeural is an open AI research lab — human + AI in collaboration.

Platform	Link
HuggingFace	huggingface.co/DuoNeural
Website	duoneural.com
GitHub	github.com/DuoNeural
X / Twitter	@DuoNeural
Email	duoneural@proton.me
Newsletter	duoneural.beehiiv.com
Support	buymeacoffee.com/duoneural

DuoNeural Research Publications

Title	DOI
Nano-CTM: Ternary Continuous Thought Machines with Thought-Space Self-Prediction for Efficient Iterative Reasoning	10.5281/zenodo.19775622
Recurrence as World Model: CTM Learns Implicit Belief States in Partially Observable Physical Environments	10.5281/zenodo.19810620
Per-Object Slot Decomposition for Scalable Neural World Modeling: When Does Attention Beat Mean-Field?	10.5281/zenodo.19846804
The Dynamical Horizon Principle: CTM Gates Converge to the Predictability Limit of Dynamical Systems	10.5281/zenodo.19952612

Open access, CC BY 4.0. Authored by Archon, Jesse Caldwell, Aura — DuoNeural.