🇺🇦 Gemma-UA-Cardio-Q4KM (Specialized Ukrainian LLM)
🩺 High-Quality Cardiology Assistant (Ukrainian Language)
This model is a highly specialized, instruction-following version of the Gemma-4B-It model (based on the size of 2.4GB, likely the 8B or a heavily compressed version), meticulously fine-tuned for providing cardiology-related information and answering medical queries in Ukrainian.
The adaptation process involved a crucial two-stage LoRA fine-tuning approach:
- Linguistic Adaptation on a large general Ukrainian corpus (14500 cardiological epicrises).
- Domain Specialization on a dedicated corpus of cardiovascular health and clinical data.
💾 Model Details & Files
This repository contains the highly optimized GGUF file, ready for immediate, efficient inference on consumer hardware.
| Detail | Value |
|---|---|
| Base Model | google/gemma-4b-it |
| Language | Ukrainian (Specialized) |
| Specialization | Cardiology (Cardiovascular medicine, clinical terminology) |
| Quantization | GGUF Q4KM (Highly efficient) |
| GGUF File Size | ~2.4 GB |
| Context Length | 4096 (Recommended minimum) |
| Pipeline Tool | llama.cpp for GGUF conversion/quantization |
Downloadable Files
gemma_ua_med_final_q8.gguf: The main file, ready for use with llama.cpp, Ollama, or LM Studio.
🇺🇦 Gemma-UA-Cardio-Q4KM (Specialized Ukrainian LLM)
🩺 High-Quality Cardiology Assistant (Ukrainian Language)
This model is a highly specialized, instruction-following version of the Gemma-4B-Instruct model (based on the size of 2.4GB, likely the 4B or a heavily compressed version), meticulously fine-tuned for providing cardiology-related information and answering medical queries in Ukrainian.
The adaptation process involved a crucial two-stage LoRA fine-tuning approach:
- Linguistic Adaptation on a large general Ukrainian corpus.
- Domain Specialization on a dedicated corpus of cardiovascular health and clinical data.
💾 Model Details & Files
This repository contains the highly optimized GGUF file, ready for immediate, efficient inference on consumer hardware.
| Detail | Value |
|---|---|
| Base Model | google/gemma-4b-it |
| Language | Ukrainian (Specialized) |
| Specialization | Cardiology (Cardiovascular medicine, clinical terminology) |
| Quantization | GGUF Q4KM (Highly efficient) |
| GGUF File Size | ~2.4 GB |
| Context Length | 4096 (Recommended minimum) |
| Pipeline Tool | llama.cpp for GGUF conversion/quantization |
Downloadable Files
gemma_ua_med_final_q4km.gguf: The main file, ready for use with llama.cpp, Ollama, or LM Studio.
🚀 How to Run (via llama.cpp CLI)
The GGUF format is ideal for running on CPU-dominant systems or systems with smaller VRAM using the llama.cpp framework.
1. Prerequisites
Ensure you have llama.cpp compiled. You will use the llama-cli binary.
2. Command Line Interface (CLI)
Use the following command to run the model in interactive chat mode. The System Prompt (-sys) is essential for activating the cardiology persona.
3. Example Usage (Ukrainian) This command assumes you are running from Ollama (Modelfile)
FROM "./gemma_ua_med_final_q4km.gguf"
SYSTEM """ Ви — лікар-кардіолог. На основі даних пацієнта та виписного епікризу сформуйте клінічно обґрунтовані рекомендації. """
TEMPLATE """<start_of_turn>user {{ .System }}
{{ .Prompt }}<end_of_turn> <start_of_turn>model """
--- IMPORTANT: Stop Tokens ---
PARAMETER stop "<end_of_turn>" PARAMETER stop "<start_of_turn>user" PARAMETER stop "<start_of_turn>model" PARAMETER stop "<start_of_turn>system"
--- Recommended generation parameters ---
PARAMETER temperature 0.6 PARAMETER top_p 0.9 PARAMETER num_ctx 4096 PARAMETER repeat_penalty 1.15 PARAMETER repeat_last_n 256
Key Parameters: -m: Specifies the path to the GGUF model file.
-sys: System Prompt—sets the model's professional role and required language.
-t $(nproc): Utilizes all available CPU cores for maximum speed.
-i: Activates interactive chat mode.
4. Limitations
- Optimized for cardiology and cardiac surgery
- Reduced accuracy outside these domains
- No vision capabilities (text-only MedGemma IT)
- May generate incomplete or generalized recommendations
5. Citing & Authors
If you use this model in your research, please cite:
@misc{Ostashko2025MedGemmaCardiology, title = {MedGemma-4B-Cardiology: A Domain-Finetuned Clinical LLM for Cardiology}, author = {Uaritm}, year = {2025}, url = {ai.esemi.org} }
Project homepage: https://ai.esemi.org
LicenseThe use of this model is subject to the terms of the original Gemma License. Please review and adhere to the associated licensing terms for the base model.