AdaptLLM/biomed-gemma-3-4b-it
AdaptLLM/biomed-gemma-3-4b-it is a 4.3 billion parameter multimodal large language model developed by AdaptLLM, fine-tuned from google/gemma-3-4b-it. This model is specifically adapted for biomedicine, integrating visual instructions for domain-specific applications. It excels at multimodal tasks within the biomedical field, leveraging its 32768 token context length for comprehensive understanding.
Loading preview...
AdaptLLM/biomed-gemma-3-4b-it: A Domain-Adapted Multimodal LLM
AdaptLLM/biomed-gemma-3-4b-it is a 4.3 billion parameter multimodal large language model (MLLM) developed by AdaptLLM, specifically adapted for the biomedicine domain. It is post-trained from the google/gemma-3-4b-it base model, as detailed in the EMNLP 2025 paper "On Domain-Adaptive Post-Training for Multimodal Large Language Models." The model integrates visual instructions, making it suitable for tasks requiring both text and image understanding in a biomedical context.
Key Capabilities and Features
- Biomedical Multimodal Understanding: Specialized for processing and generating responses based on biomedical images and text.
- Instruction-Tuned: Designed to follow instructions effectively, particularly when images are provided at the beginning of the input.
- Gemma-3-4b-it Architecture: Leverages the robust architecture of the Gemma-3-4b-it model, ensuring strong foundational language capabilities.
- Reproducible Training: The model's training process, using the biomed-visual-instructions dataset and LlamaFactory, is openly documented for reproducibility.
Use Cases and Evaluation
This model is ideal for applications requiring multimodal reasoning in biomedicine, such as visual question answering (VQA) on medical images. AdaptLLM provides a dedicated biomed-VQA-benchmark for evaluating MLLMs in this domain. Developers can easily integrate the model using the Hugging Face pipeline API, with examples provided for chat-based interactions.