Medgemma-4B-NPU2

Instructions to use OpenFlowLM/Medgemma-4B-NPU2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use OpenFlowLM/Medgemma-4B-NPU2 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="OpenFlowLM/Medgemma-4B-NPU2")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForMultimodalLM

tokenizer = AutoTokenizer.from_pretrained("OpenFlowLM/Medgemma-4B-NPU2")
model = AutoModelForMultimodalLM.from_pretrained("OpenFlowLM/Medgemma-4B-NPU2", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use OpenFlowLM/Medgemma-4B-NPU2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "OpenFlowLM/Medgemma-4B-NPU2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "OpenFlowLM/Medgemma-4B-NPU2",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/OpenFlowLM/Medgemma-4B-NPU2

SGLang

How to use OpenFlowLM/Medgemma-4B-NPU2 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "OpenFlowLM/Medgemma-4B-NPU2" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "OpenFlowLM/Medgemma-4B-NPU2",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "OpenFlowLM/Medgemma-4B-NPU2" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "OpenFlowLM/Medgemma-4B-NPU2",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use OpenFlowLM/Medgemma-4B-NPU2 with Docker Model Runner:
```
docker model run hf.co/OpenFlowLM/Medgemma-4B-NPU2
```

Medgemma-4B-NPU2 / README.md

Atomic-Germ

Duplicate from Atomic-Germ/Medgemma-4B-NPU2

3ecb5f0 20 days ago

preview code

Raw History Blame Contribute Delete

2.68 kB

---
license: other
base_model:
- Atomic-Germ/Medgemma-4B-NPU2
base_model_relation: quantized
quantized_by: Atomic-Germ
pipeline_tag: image-text-to-text
oflm-family: gemma3
tags:
- transformers
- gemma3
- image-text-to-text
- medical
- radiology
- clinical-reasoning
- dermatology
- pathology
- ophthalmology
- chest-x-ray
- arxiv:2303.15343
- arxiv:2507.05201
- arxiv:2405.03162
- arxiv:2106.14463
- arxiv:2412.03555
- arxiv:2501.19393
- arxiv:2009.13081
- arxiv:2102.09542
- arxiv:2411.15640
- arxiv:2404.05590
- arxiv:2501.18362
- base_model:google/gemma-3-4b-pt
- base_model:finetune:google/gemma-3-4b-pt
- license:other
- endpoints_compatible
- region:us
- oflm
- openflowlm
- npu2
- q4nx
---
# Medgemma-4B-NPU2

**OpenFlowLM Q4NX conversion of [`Atomic-Germ/Medgemma-4B-NPU2`](https://huggingface.co/Atomic-Germ/Medgemma-4B-NPU2)** for AMD XDNA NPU inference.

This repository contains a quantized **Q4NX** port of the model, compiled for the OpenFlowLM (OFLM) runtime. It is **not** a GGUF file.

| Item | Value |
|------|-------|
| Source model | [`Atomic-Germ/Medgemma-4B-NPU2`](https://huggingface.co/Atomic-Germ/Medgemma-4B-NPU2) |
| Source GGUF | `medgemma-4b-it.i1-Q4_1.gguf` |
| Weights | `model.q4nx` (3.51 GB) |
| Modality | language |
| OFLM version | `0.1.0` |
| Converted | 2026-10-01 |

## Reproduce this conversion

This container was produced by exactly this command, so a fine-tune of the
same base can be packed the same way:

```bash
oflm pack -i /home/atomic-germ/.cache/huggingface/hub/models--mradermacher--medgemma-4b-it-i1-GGUF/snapshots/e72473b68139c8f77427dd21270eff3b2b80c1f9/medgemma-4b-it.i1-Q4_1.gguf -o /home/atomic-germ/Projects/OpenFlowLM-Next/Medgemma-4B-NPU2 -s Atomic-Germ/Medgemma-4B-NPU2
```


## Install and run

This repository works with `oflm-add`, a small installer that copies the model
into the OpenFlowLM user directory and registers the tag. It never
modifies the system OpenFlowLM install.

`pip install oflm-add` or `uv tool install oflm-add`

```bash
uv tool install oflm-add
oflm-add Atomic-Germ/Medgemma-4B-NPU2 --family qwen3.5 --xclbin-from Medgemma-4B-NPU2
OFLM_CONFIG_PATH="$HOME/.config/oflm/model_list.json" OFLM_XCLBIN_PATH="$HOME/.config/oflm" oflm run Medgemma-4B-NPU2
```

## Files

| File | Description |
|------|-------------|
| `model.q4nx` | Quantized weights (Q8_0 / Q4_1 / BF16) |
| `config.json` | OFLM runtime configuration |
| `tokenizer.json` | Tokenizer vocabulary |
| `tokenizer_config.json` | Tokenizer configuration |
| `chat_template.jinja` | Chat template |


---

## Source model card

See the original model card: [Atomic-Germ/Medgemma-4B-NPU2](https://huggingface.co/Atomic-Germ/Medgemma-4B-NPU2)

	---
	license: other
	base_model:
	- Atomic-Germ/Medgemma-4B-NPU2
	base_model_relation: quantized
	quantized_by: Atomic-Germ
	pipeline_tag: image-text-to-text
	oflm-family: gemma3
	tags:
	- transformers
	- gemma3
	- image-text-to-text
	- medical
	- radiology
	- clinical-reasoning
	- dermatology
	- pathology
	- ophthalmology
	- chest-x-ray
	- arxiv:2303.15343
	- arxiv:2507.05201
	- arxiv:2405.03162
	- arxiv:2106.14463
	- arxiv:2412.03555
	- arxiv:2501.19393
	- arxiv:2009.13081
	- arxiv:2102.09542
	- arxiv:2411.15640
	- arxiv:2404.05590
	- arxiv:2501.18362
	- base_model:google/gemma-3-4b-pt
	- base_model:finetune:google/gemma-3-4b-pt
	- license:other
	- endpoints_compatible
	- region:us
	- oflm
	- openflowlm
	- npu2
	- q4nx
	---
	# Medgemma-4B-NPU2

	OpenFlowLM Q4NX conversion of [`Atomic-Germ/Medgemma-4B-NPU2`](https://huggingface.co/Atomic-Germ/Medgemma-4B-NPU2) for AMD XDNA NPU inference.

	This repository contains a quantized Q4NX port of the model, compiled for the OpenFlowLM (OFLM) runtime. It is not a GGUF file.

	\| Item \| Value \|
	\|------\|-------\|
	\| Source model \| [`Atomic-Germ/Medgemma-4B-NPU2`](https://huggingface.co/Atomic-Germ/Medgemma-4B-NPU2) \|
	\| Source GGUF \| `medgemma-4b-it.i1-Q4_1.gguf` \|
	\| Weights \| `model.q4nx` (3.51 GB) \|
	\| Modality \| language \|
	\| OFLM version \| `0.1.0` \|
	\| Converted \| 2026-10-01 \|

	## Reproduce this conversion

	This container was produced by exactly this command, so a fine-tune of the
	same base can be packed the same way:

	```bash
	oflm pack -i /home/atomic-germ/.cache/huggingface/hub/models--mradermacher--medgemma-4b-it-i1-GGUF/snapshots/e72473b68139c8f77427dd21270eff3b2b80c1f9/medgemma-4b-it.i1-Q4_1.gguf -o /home/atomic-germ/Projects/OpenFlowLM-Next/Medgemma-4B-NPU2 -s Atomic-Germ/Medgemma-4B-NPU2
	```


	## Install and run

	This repository works with `oflm-add`, a small installer that copies the model
	into the OpenFlowLM user directory and registers the tag. It never
	modifies the system OpenFlowLM install.

	`pip install oflm-add` or `uv tool install oflm-add`

	```bash
	uv tool install oflm-add
	oflm-add Atomic-Germ/Medgemma-4B-NPU2 --family qwen3.5 --xclbin-from Medgemma-4B-NPU2
	OFLM_CONFIG_PATH="$HOME/.config/oflm/model_list.json" OFLM_XCLBIN_PATH="$HOME/.config/oflm" oflm run Medgemma-4B-NPU2
	```

	## Files

	\| File \| Description \|
	\|------\|-------------\|
	\| `model.q4nx` \| Quantized weights (Q8_0 / Q4_1 / BF16) \|
	\| `config.json` \| OFLM runtime configuration \|
	\| `tokenizer.json` \| Tokenizer vocabulary \|
	\| `tokenizer_config.json` \| Tokenizer configuration \|
	\| `chat_template.jinja` \| Chat template \|


	---

	## Source model card

	See the original model card: [Atomic-Germ/Medgemma-4B-NPU2](https://huggingface.co/Atomic-Germ/Medgemma-4B-NPU2)