Downloads · 30 days
116
12% of all-time downloads
naazimsnh02/dentalgemma-1.5-4b-it-GGUF
dentalgemma-1.5-4b-it-GGUF is a image-text-to-text model from naazimsnh02. Use it for the image-text-to-text task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
This repository contains the GGUF quantized formats of DentalGemma 1.5 4B IT. These files are designed for efficient local inference using llama.cpp and compatible frameworks on edge devices, mobile applications (such…
Downloads · 30 days
116
12% of all-time downloads
All-time downloads
935
Public
Repo size
3.7 GB
Likes
0
Public
Click a slice to open those files.
.gguf3.7 GB · 100%
From the Hugging Face model README
This repository contains the GGUF quantized formats of DentalGemma 1.5 4B IT. These files are designed for efficient local inference using llama.cpp and compatible frameworks on edge devices, mobile applications (such as React Native with llama.rn), and consumer hardware.
DentalGemma is a fine-tuned version of MedGemma 1.5 4B IT specialized for dental diagnostics and clinical reasoning. It represents a novel domain adaptation, bridging the gap between medical foundation models and localized dental expertise.
Built for the MedGemma Impact Challenge, this GGUF version allows the DentalGemma model to be brought directly into the hands of clinicians and patients via affordable low-resource devices and mobile platforms, entirely offline and privacy-preserving.
DentalGemma is a multimodal vision-language model that combines:
For full details on the training data, methodology, benchmark performance, and full precision evaluation, please refer to the original model card.
The GGUF format enables highly efficient CPU and GPU inference without requiring heavy Python dependencies, PyTorch, or Transformers.
llama.cpp following the instructions in the llama.cpp repository.mmproj-*.gguf) and the main quantized model file (*.gguf) from the files section.Use the llama-llava-cli tool (or llava-cli depending on your llama.cpp version/build) to process images and text simultaneously:
./llama-llava-cli -m dentalgemma-1.5-4b-it.Q4_K_M.gguf \
--mmproj mmproj-dentalgemma-1.5-4b-it-f16.gguf \
-p "<bos><start_of_turn>user\n[img-1]Analyze this dental X-ray for any abnormalities.<end_of_turn>\n<start_of_turn>model\n" \
--image dental_xray.jpg \
-c 4096 \
--temp 0.1
Use the standard llama-cli tool for clinical case reasoning and textual interactions:
./llama-cli -m dentalgemma-1.5-4b-it.Q4_K_M.gguf \
-p "<bos><start_of_turn>system\nYou are an expert dental clinician and radiologist AI assistant.<end_of_turn>\n<start_of_turn>user\nA 35-year-old male presents with severe throbbing pain in the lower right molar region for 3 days. What is your assessment and management plan?<end_of_turn>\n<start_of_turn>model\n" \
-c 4096 \
--temp 0.1
This GGUF model powers the offline-first dentalgemma-mobile React Native app via llama.rn integration. When deploying to mobile devices:
.gguf) and multimodal projector weights (mmproj.gguf) to the iOS/Android device's local file storage.Q4_K_M) requires approximately ~2.5GB to 3GB RAM, making it feasible for execution on mid-to-high-end smartphones (8GB+ unified memory recommended for multimodal generation).llama.cpp K-quants (e.g., Q4_K_M, Q8_0)The base model was first fine-tuned in full bfloat16 using LoRA. The learned adapters were then merged into the base weights, and the resulting full-precision model was converted to GGUF format using llama.cpp scripts (convert_image_encoder_to_gguf.py for the projector and convert_hf_to_gguf.py for the LLM).
bfloat16 weights. We recommend Q4_K_M or higher for the best balance of speed and accuracy.This model is NOT a substitute for professional dental diagnosis. All AI-generated assessments must be validated by licensed dental professionals before any clinical use. The model is provided "as-is" for research purposes only.
If you use DentalGemma in your research, please cite:
@misc{dentalgemma2026,
title={DentalGemma: Fine-tuning MedGemma for Dental Diagnostics},
author={Syed Naazim Hussain},
year={2026},
publisher={HuggingFace},
howpublished={\url{https://huggingface.co/naazimsnh02/dentalgemma-1.5-4b-it-GGUF}},
note={MedGemma Impact Challenge submission}
}
Built for the MedGemma Impact Challenge 🏥
Bringing dental diagnostics into the age of medical foundation models 🦷
</div>