Quick start with the MedGemma-4b model

MedGemma-4b is a collection of Gemma 3 variants that are trained for performance on medical text and image comprehension. Developers can use MedGemma to accelerate building healthcare-based AI applications. MedGemma currently comes in two variants: a 4B multimodal version and a 27B text-only version.

Step 1: Install WasmEdge

First off, you'll need WasmEdge, a high-performance, lightweight, and cross-platform LLM runtime.

curl -sSf https://raw.githubusercontent.com/WasmEdge/WasmEdge/master/utils/install_v2.sh | bash -s

Step 2: Download the LLM model

Next, you'll need to obtain model files: the MedGemma model and the mmproj model.

curl -LO https://huggingface.co/second-state/medgemma-4b-it-GGUF/resolve/main/medgemma-4b-it-Q5_K_M.gguf
curl -LO https://huggingface.co/second-state/medgemma-4b-it-GGUF/resolve/main/medgemma-4b-it-mmproj-f16.gguf

Step 3: Download a portable API server app

Next, you need an application that can build and OpenAI-compatible API server for the MedGemma models The LlamaEdge api server app is a lightweight and cross-platform Wasm app that works on any device you might have. Just download the compiled binary app.

curl -LO https://github.com/second-state/LlamaEdge/releases/latest/download/llama-api-server.wasm

The version of the llama-api-server.wasm should be v0.18.5 and above.

Step 4: Chat with the chatbot UI

The llama-api-server.wasm is a web server with an OpenAI-compatible API. You still need HTML files for the chatbot UI. It's optional and you can use curl to send an API request. Download and unzip the HTML UI files as follows.

curl -LO https://github.com/LlamaEdge/chatbot-ui/releases/latest/download/chatbot-ui.tar.gz
tar xzf chatbot-ui.tar.gz
rm chatbot-ui.tar.gz

Then, start the web server.

wasmedge --dir .:. --nn-preload default:GGML:AUTO:medgemma-4b-it-Q5_K_M.gguf \
  llama-api-server.wasm \
  --prompt-template gemma-3 \
  --llava-mmproj medgemma-4b-it-mmproj-f16.gguf \
  --ctx-size 4098 \
  --model-name medgemma-4b

The above command lines work on a Macbook with 16 GB memory.

Upon successful execution, you should see output similar to the following:

[2025-05-29 17:07:46.398] [info] llama_api_server in llama-api-server/src/main.rs:544: model_name: medgemma-4b
[2025-05-29 17:07:46.398] [info] llama_api_server in llama-api-server/src/main.rs:553: model_alias: default
[2025-05-29 17:07:46.398] [info] llama_api_server in llama-api-server/src/main.rs:573: ctx_size: 4098
[2025-05-29 17:07:46.398] [info] llama_api_server in llama-api-server/src/main.rs:593: batch_size: 512

...
[2025-05-29 17:07:46.935] [info] llama_api_server in llama-api-server/src/main.rs:907: running_mode: chat
[2025-05-29 17:07:46.935] [info] llama_api_server in llama-api-server/src/main.rs:917: plugin_ggml_version: b5201 (commit 85f36e5e)
[2025-05-29 17:07:46.936] [info] llama_api_server in llama-api-server/src/main.rs:952: Listening on 0.0.0.0:8080

Then, go to http://localhost:8080 on your computer to access the chatbot UI on a web page! You can upload an imange and chat with the model based on the medical image.

Step 5: Send an API request

You can send an API request to call the model, which is more universal. The following command demonstrates how to send a CURL request to llama-api-server. The request includes a base64-encoded string of an image in the image_url field. For demonstration purposes, only a portion of the base64 string is shown here. In practice, you should use the complete base64 string.

[!TIP] base64.guru provides a tool for encoding JPG to Base64.

curl --location 'http://localhost:8080/v1/chat/completions' \
--header 'Content-Type: application/json' \
--data '{
    "messages": [
        {
            "role": "system",
            "content": "You are a helpful medical assistant that accurately describes the content of images provided by the user."
        },
        {
            "content": [
                {
                    "type": "text",
                    "text": "Read this chest X-ray and find anything abnormal"
                },
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAgFBgcGBQg......X/VaTer/ALzOU/Lg1XMiLMuR3EWMb77/AHsD/DNTIhXPnmvLmwj"
                    }
                }
            ],
            "role": "user"
        }
    ],
    "model": "medgemma-4b"
}'

If the request is processed successfully, you will receive a response similar to the following:

{
    "id": "chatcmpl-e5f777db-c913-45ab-b37f-e2c499c8fa0b",
    "object": "chat.completion",
    "created": 1747652210,
    "model": "medgemma-4b",
    "choices": [
        {
            "index": 0,
            "message": {
                "content": "There is a round, dense opacity in the right lower lobe of the lung. This could be a mass or nodule, and further investigation would be needed to determine its nature.",
                "role": "assistant"
            },
            "finish_reason": "stop",
            "logprobs": null
        }
    ],
    "usage": {
        "prompt_tokens": 27,
        "completion_tokens": 68,
        "total_tokens": 95
    }
}

You can also combine the base64 conversion and API call into a single command like this:

curl --location 'http://localhost:8080/v1/chat/completions' \
--header 'Content-Type: application/json' \
--data '{"messages": [{"role": "system","content": "You are a helpful medical assistant that accurately describes the content of images provided by the user."}, {"content": [{"type": "text","text": ""Read this chest X-ray and find anything abnormal"},{"type": "image_url","image_url": {"url": "'"$(base64 -i /path/to/image.jpg)"'"}}], "role": "user"}], "model": "medgemma-4b"}'

Congratulations! You have now started an multimodal app on your own device.