Instructions to use bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4A16 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4A16 with Transformers:

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

pipe = pipeline("text-generation", model="bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4A16")
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)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4A16")
model = AutoModelForMultimodalLM.from_pretrained("bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4A16")
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 = processor.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=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4A16 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4A16"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4A16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4A16

SGLang

How to use bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4A16 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 "bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4A16" \
    --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": "bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4A16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4A16" \
        --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": "bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4A16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4A16 with Docker Model Runner:
```
docker model run hf.co/bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4A16
```

marioiseli commited on Apr 3

Commit

c4c83dd

verified ·

1 Parent(s): 77cc23d

Upload gemma4_patched.py with huggingface_hub

Browse files

Files changed (1) hide show

gemma4_patched.py +0 -5

gemma4_patched.py CHANGED Viewed

@@ -1027,9 +1027,6 @@ class Gemma4Model(nn.Module):
                     if is_pp_missing_parameter(moe_name, self):
                         continue
                     param = params_dict[moe_name]
-                    # Debug: log first expert of first layer
-                    if expert_id == 0 and "layers.0." in name:
-                        print(f"[PATCH LOAD] {name} -> {moe_name} | wl={wl_name} shard={shard_id} eid={expert_id} | weight:{list(loaded_weight.shape)} param:{list(param.shape)}")
                     weight_loader = param.weight_loader
                     weight_loader(
                         param,
@@ -1049,7 +1046,6 @@ class Gemma4Model(nn.Module):
                     if is_pp_missing_parameter(name, self):
                         continue
                     if name not in params_dict:
-                        print(f"[PATCH DEBUG] Skipping unknown key: {name}")
                         continue
                     param = params_dict[name]
                     weight_loader = getattr(
@@ -1060,7 +1056,6 @@ class Gemma4Model(nn.Module):
                     import inspect
                     sig = inspect.signature(weight_loader)
                     if len(sig.parameters) > 3 and "expert_id" in sig.parameters:
-                        print(f"[PATCH DEBUG] Skipping FusedMoE param that fell through mapping: {name}")
                         continue
                     weight_loader(param, loaded_weight)
             loaded_params.add(name)

                     if is_pp_missing_parameter(moe_name, self):
                         continue
                     param = params_dict[moe_name]
                     weight_loader = param.weight_loader
                     weight_loader(
                         param,
                     if is_pp_missing_parameter(name, self):
                         continue
                     if name not in params_dict:
                         continue
                     param = params_dict[name]
                     weight_loader = getattr(
                     import inspect
                     sig = inspect.signature(weight_loader)
                     if len(sig.parameters) > 3 and "expert_id" in sig.parameters:
                         continue
                     weight_loader(param, loaded_weight)
             loaded_params.add(name)