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

Libraries

How to use unsloth/granite-4.0-h-tiny-base with Transformers:

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

pipe = pipeline("text-generation", model="unsloth/granite-4.0-h-tiny-base")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("unsloth/granite-4.0-h-tiny-base")
model = AutoModelForCausalLM.from_pretrained("unsloth/granite-4.0-h-tiny-base")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use unsloth/granite-4.0-h-tiny-base with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/unsloth/granite-4.0-h-tiny-base

SGLang

How to use unsloth/granite-4.0-h-tiny-base 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 "unsloth/granite-4.0-h-tiny-base" \
    --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": "unsloth/granite-4.0-h-tiny-base",
		"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 "unsloth/granite-4.0-h-tiny-base" \
        --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": "unsloth/granite-4.0-h-tiny-base",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Unsloth Studio new

How to use unsloth/granite-4.0-h-tiny-base with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for unsloth/granite-4.0-h-tiny-base to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for unsloth/granite-4.0-h-tiny-base to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for unsloth/granite-4.0-h-tiny-base to start chatting

Load model with FastModel

pip install unsloth
from unsloth import FastModel
model, tokenizer = FastModel.from_pretrained(
    model_name="unsloth/granite-4.0-h-tiny-base",
    max_seq_length=2048,
)

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

granite-4.0-h-tiny-base / config.json

danielhanchen

Upload folder using huggingface_hub

ab0e551 verified 7 months ago

raw

history blame contribute delete

1.82 kB

	{
	"architectures": [
	"GraniteMoeHybridForCausalLM"
	],
	"attention_bias": false,
	"attention_dropout": 0.0,
	"attention_multiplier": 0.0078125,
	"bos_token_id": 100257,
	"torch_dtype": "bfloat16",
	"embedding_multiplier": 12,
	"eos_token_id": 100257,
	"hidden_act": "silu",
	"hidden_size": 1536,
	"initializer_range": 0.1,
	"intermediate_size": 512,
	"layer_types": [
	"mamba",
	"mamba",
	"mamba",
	"mamba",
	"mamba",
	"attention",
	"mamba",
	"mamba",
	"mamba",
	"mamba",
	"mamba",
	"mamba",
	"mamba",
	"mamba",
	"mamba",
	"attention",
	"mamba",
	"mamba",
	"mamba",
	"mamba",
	"mamba",
	"mamba",
	"mamba",
	"mamba",
	"mamba",
	"attention",
	"mamba",
	"mamba",
	"mamba",
	"mamba",
	"mamba",
	"mamba",
	"mamba",
	"mamba",
	"mamba",
	"attention",
	"mamba",
	"mamba",
	"mamba",
	"mamba"
	],
	"logits_scaling": 6,
	"mamba_chunk_size": 256,
	"mamba_conv_bias": true,
	"mamba_d_conv": 4,
	"mamba_d_head": 64,
	"mamba_d_state": 128,
	"mamba_expand": 2,
	"mamba_n_groups": 1,
	"mamba_n_heads": 48,
	"mamba_proj_bias": false,
	"max_position_embeddings": 131072,
	"model_type": "granitemoehybrid",
	"normalization_function": "rmsnorm",
	"num_attention_heads": 12,
	"num_experts_per_tok": 6,
	"num_hidden_layers": 40,
	"num_key_value_heads": 4,
	"num_local_experts": 64,
	"output_router_logits": false,
	"pad_token_id": 100256,
	"position_embedding_type": "nope",
	"residual_multiplier": 0.22,
	"rms_norm_eps": 1e-05,
	"rope_scaling": null,
	"rope_theta": 10000,
	"router_aux_loss_coef": 0.0,
	"shared_intermediate_size": 1024,
	"tie_word_embeddings": true,
	"transformers_version": "4.57.1",
	"unsloth_fixed": true,
	"use_cache": true,
	"vocab_size": 100352
	}