Instructions to use ibm-granite/granite-4.2-3b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ibm-granite/granite-4.2-3b with Transformers:

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

pipe = pipeline("text-generation", model="ibm-granite/granite-4.2-3b")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("ibm-granite/granite-4.2-3b")
model = AutoModelForCausalLM.from_pretrained("ibm-granite/granite-4.2-3b", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Inference
HuggingChat
Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use ibm-granite/granite-4.2-3b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ibm-granite/granite-4.2-3b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ibm-granite/granite-4.2-3b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/ibm-granite/granite-4.2-3b

SGLang

How to use ibm-granite/granite-4.2-3b 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-4.2-3b" \
    --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": "ibm-granite/granite-4.2-3b",
		"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 "ibm-granite/granite-4.2-3b" \
        --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": "ibm-granite/granite-4.2-3b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use ibm-granite/granite-4.2-3b with Docker Model Runner:
```
docker model run hf.co/ibm-granite/granite-4.2-3b
```

Clarification on enabling and evaluating 512K context

by pengwenzhi - opened 9 days ago

Discussion

pengwenzhi

9 days ago

Hi Granite team,

The model card says Granite-4.2-3B natively supports 128K, with long-context extension to 512K, but the architecture still lists a sequence length of 131072, and the vLLM/SGLang examples do not show how to enable 512K.

Could you clarify the recommended configuration for 512K inference?

Is 512K officially supported by the released checkpoint?
Does it require any RoPE scaling or other positional-encoding settings?
For vLLM, is --max-model-len 524288 sufficient?
For SGLang, is --context-length 524288 sufficient?
Do you have an official 512K evaluation or launch example?

Thanks!

gabegoodhart

IBM Granite org 8 days ago

Hi @pengwenzhi thanks for the question! Here are the details on the 128k vs 512k context lengths:

All models in the 4.2 family were trained to support 512k context, so for each inference platform, you can simply use the corresponding flag to raise the max context length (no RoPE scaling needed)
All models in the 4.2 were benchmarked and evaluated using 128k context, so the default values for each distributed checkpoint format is set to 128k as the maximum verified length

pengwenzhi

5 days ago

I got it.

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