Instructions to use ibm-granite/granite-4.0-3b-vision with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ibm-granite/granite-4.0-3b-vision with Transformers:

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

pipe = pipeline("image-text-to-text", model="ibm-granite/granite-4.0-3b-vision")

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("ibm-granite/granite-4.0-3b-vision", dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use ibm-granite/granite-4.0-3b-vision with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ibm-granite/granite-4.0-3b-vision"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ibm-granite/granite-4.0-3b-vision",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/ibm-granite/granite-4.0-3b-vision

SGLang

How to use ibm-granite/granite-4.0-3b-vision 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 "ibm-granite/granite-4.0-3b-vision" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ibm-granite/granite-4.0-3b-vision",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "ibm-granite/granite-4.0-3b-vision" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ibm-granite/granite-4.0-3b-vision",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use ibm-granite/granite-4.0-3b-vision with Docker Model Runner:
```
docker model run hf.co/ibm-granite/granite-4.0-3b-vision
```

Docker error The size of tensor a (1280) must match the size of tensor b (2560) at non-singleton dimension 1

by kuopching - opened Apr 7

Discussion

kuopching

Apr 7

Hi, i have trouble to run ibm-granite/granite-4.0-3b-vision using docker.

vllm version: 0.19.0
command:
docker run --gpus all -p 9999:8000 --ipc=host vllm-granite-vision --model ibm-granite/granite-4.0-3b-vision --trust-remote-code --max-model-len 16384 --tens
or-parallel-size 2 --host 0.0.0.0 --port 8000 --hf-overrides '{"adapter_path": "ibm-granite/granite-4.0-3b-vision"}' --gpu-memory-utilization 0.7

error:
(Worker_TP1 pid=83) ERROR 04-07 07:26:04 [multiproc_executor.py:857] loaded_weights = model.load_weights(self.get_all_weights(model_config, model))
(Worker_TP1 pid=83) ERROR 04-07 07:26:04 [multiproc_executor.py:857] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP1 pid=83) ERROR 04-07 07:26:04 [multiproc_executor.py:857] File "/app/granite4_vision.py", line 948, in load_weights
(Worker_TP1 pid=83) ERROR 04-07 07:26:04 [multiproc_executor.py:857] self._apply_adapter()
(Worker_TP1 pid=83) ERROR 04-07 07:26:04 [multiproc_executor.py:857] File "/app/granite4_vision.py", line 940, in _apply_adapter
(Worker_TP1 pid=83) ERROR 04-07 07:26:04 [multiproc_executor.py:857] n = self._merge_lora_deltas(adapter_config, adapter_weights)
(Worker_TP1 pid=83) ERROR 04-07 07:26:04 [multiproc_executor.py:857] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP1 pid=83) ERROR 04-07 07:26:04 [multiproc_executor.py:857] File "/app/granite4_vision.py", line 919, in _merge_lora_deltas
(Worker_TP1 pid=83) ERROR 04-07 07:26:04 [multiproc_executor.py:857] if _add_delta(module_key + ".weight", delta):
(Worker_TP1 pid=83) ERROR 04-07 07:26:04 [multiproc_executor.py:857] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP1 pid=83) ERROR 04-07 07:26:04 [multiproc_executor.py:857] File "/app/granite4_vision.py", line 906, in _add_delta
(Worker_TP1 pid=83) ERROR 04-07 07:26:04 [multiproc_executor.py:857] param.data = (param.data.float() + delta.to(param.device)).to(param.dtype)
(Worker_TP1 pid=83) ERROR 04-07 07:26:04 [multiproc_executor.py:857] ~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~
(Worker_TP1 pid=83) ERROR 04-07 07:26:04 [multiproc_executor.py:857] RuntimeError: The size of tensor a (1280) must match the size of tensor b (2560) at non-singleton dimension 1

artemspector

IBM Granite org Apr 9

Hi, there might be an issue with handling tensor parallelism in case of full merge flow. I'm going to check it.
Meanwhile you can try the inference on a single GPU by removing the command arg --tensor-parallel-size 2

artemspector

IBM Granite org Apr 12

Pushed the fixed version, should work now with multiple GPUs

artemspector changed discussion status to closed Apr 12

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment