Instructions to use Open4bits/granite-4.0-h-tiny-f16 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Open4bits/granite-4.0-h-tiny-f16 with Transformers:

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

pipe = pipeline("text-generation", model="Open4bits/granite-4.0-h-tiny-f16")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Open4bits/granite-4.0-h-tiny-f16")
model = AutoModelForCausalLM.from_pretrained("Open4bits/granite-4.0-h-tiny-f16")
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]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Open4bits/granite-4.0-h-tiny-f16 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Open4bits/granite-4.0-h-tiny-f16"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Open4bits/granite-4.0-h-tiny-f16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Open4bits/granite-4.0-h-tiny-f16

SGLang

How to use Open4bits/granite-4.0-h-tiny-f16 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 "Open4bits/granite-4.0-h-tiny-f16" \
    --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": "Open4bits/granite-4.0-h-tiny-f16",
		"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 "Open4bits/granite-4.0-h-tiny-f16" \
        --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": "Open4bits/granite-4.0-h-tiny-f16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Open4bits/granite-4.0-h-tiny-f16 with Docker Model Runner:
```
docker model run hf.co/Open4bits/granite-4.0-h-tiny-f16
```

Open4bits / Granite 4.0 H Tiny FP16

This repository provides the Granite 4.0 H Tiny model converted to FP16 (float16) precision, published by Open4bits for more efficient inference while preserving the original model behavior.

The underlying Granite model architecture and weights are owned by IBM. This repository contains only a precision-converted version of the original model.

The model is suitable for general-purpose text generation tasks and lightweight language model experimentation.

Model Overview

Granite is a family of transformer-based language models developed by IBM, designed for efficient and scalable natural language processing tasks.

This release uses the Granite 4.0 H Tiny variant, focusing on lower memory usage and faster inference while maintaining architectural fidelity to the original model.

Model Details

Architecture: Granite 4.0 H Tiny
Precision: float16 (FP16)
Task: Text Generation
Weight tying: Preserved
Compatibility: Hugging Face Transformers, PyTorch

The FP16 conversion reduces VRAM requirements and improves inference speed compared to FP32 versions, making the model suitable for deployment on resource-constrained hardware.

Intended Use

This model is intended for:

Text generation and completion
Research and experimentation
Lightweight inference workloads
Prototyping and benchmarking language models

Limitations

Inherits limitations of the Granite Tiny architecture
Output quality may be lower than larger Granite variants
Not fine-tuned for domain-specific or instruction-following tasks

License

This model is released under the Apache License 2.0. The original Granite model and associated intellectual property are owned by IBM.

Support

If you find this model useful, please consider supporting the project. Your support helps us continue releasing and maintaining high-quality open models. Support us with a heart.

Downloads last month: 1

Safetensors

Model size

7B params

Tensor type

BF16

Model tree for Open4bits/granite-4.0-h-tiny-f16

Base model

ibm-granite/granite-4.0-h-tiny

Finetuned

(9)

this model