by ibm-granite
Open source · 291k downloads · 268 likes
Granite 4.0 Micro is a lightweight AI model with 3 billion parameters, specifically designed to follow precise instructions and interact with external tools. Developed by IBM, it excels in a variety of tasks such as text summarization, classification, information extraction, question answering, and retrieval-augmented generation (RAG). With its multilingual capabilities covering 12 languages and its ability to call functions, it seamlessly integrates into virtual assistants and autonomous enterprise agents. What sets it apart is its enhanced alignment for professional, safe, and accurate responses, as well as its flexibility to be tailored to specific needs. Ideal for professional applications or deployments requiring efficient context and tool management.
📣 Update [10-07-2025]: Added a default system prompt to the chat template to guide the model towards more professional, accurate, and safe responses.
Model Summary: Granite-4.0-Micro is a 3B parameter long-context instruct model finetuned from Granite-4.0-Micro-Base using a combination of open source instruction datasets with permissive license and internally collected synthetic datasets. This model is developed using a diverse set of techniques with a structured chat format, including supervised finetuning, model alignment using reinforcement learning, and model merging. Granite 4.0 instruct models feature improved instruction following (IF) and tool-calling capabilities, making them more effective in enterprise applications.
Supported Languages: English, German, Spanish, French, Japanese, Portuguese, Arabic, Czech, Italian, Korean, Dutch, and Chinese. Users may finetune Granite 4.0 models for languages beyond these languages.
Intended use: The model is designed to follow general instructions and can serve as the foundation for AI assistants across diverse domains, including business applications, as well as for LLM agents equipped with tool-use capabilities.
Capabilities
Generation: This is a simple example of how to use Granite-4.0-Micro model.
Install the following libraries:
pip install torch torchvision torchaudio
pip install accelerate
pip install transformers
Then, copy the snippet from the section that is relevant for your use case.
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
device = "cuda"
model_path = "ibm-granite/granite-4.0-micro"
tokenizer = AutoTokenizer.from_pretrained(model_path)
# drop device_map if running on CPU
model = AutoModelForCausalLM.from_pretrained(model_path, device_map=device)
model.eval()
# change input text as desired
chat = [
{ "role": "user", "content": "Please list one IBM Research laboratory located in the United States. You should only output its name and location." },
]
chat = tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
# tokenize the text
input_tokens = tokenizer(chat, return_tensors="pt").to(device)
# generate output tokens
output = model.generate(**input_tokens,
max_new_tokens=100)
# decode output tokens into text
output = tokenizer.batch_decode(output)
# print output
print(output[0])
Expected output:
<|start_of_role|>system<|end_of_role|>You are a helpful assistant. Please ensure responses are professional, accurate, and safe.<|end_of_text|>
<|start_of_role|>user<|end_of_role|>Please list one IBM Research laboratory located in the United States. You should only output its name and location.<|end_of_text|>
<|start_of_role|>assistant<|end_of_role|>Almaden Research Center, San Jose, California<|end_of_text|>
Tool-calling: Granite-4.0-Micro comes with enhanced tool calling capabilities, enabling seamless integration with external functions and APIs. To define a list of tools please follow OpenAI's function definition schema.
This is an example of how to use Granite-4.0-Micro model tool-calling ability:
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
device = "cuda"
model_path = "ibm-granite/granite-4.0-micro"
tokenizer = AutoTokenizer.from_pretrained(model_path)
# drop device_map if running on CPU
model = AutoModelForCausalLM.from_pretrained(model_path, device_map=device)
model.eval()
tools = [
{
"type": "function",
"function": {
"name": "get_current_weather",
"description": "Get the current weather for a specified city.",
"parameters": {
"type": "object",
"properties": {
"city": {
"type": "string",
"description": "Name of the city"
}
},
"required": ["city"]
}
}
}
]
# change input text as desired
chat = [
{ "role": "user", "content": "What's the weather like in Boston right now?" },
]
chat = tokenizer.apply_chat_template(chat, \
tokenize=False, \
tools=tools, \
add_generation_prompt=True)
# tokenize the text
input_tokens = tokenizer(chat, return_tensors="pt").to(device)
# generate output tokens
output = model.generate(**input_tokens,
max_new_tokens=100)
# decode output tokens into text
output = tokenizer.batch_decode(output)
# print output
print(output[0])
Expected output:
<|start_of_role|>system<|end_of_role|>You are a helpful assistant with access to the following tools. You may call one or more tools to assist with the user query.
You are provided with function signatures within <tools></tools> XML tags:
<tools>
{"type": "function", "function": {"name": "get_current_weather", "description": "Get the current weather for a specified city.", "parameters": {"type": "object", "properties": {"city": {"type": "string", "description": "Name of the city"}}, "required": ["city"]}}}
</tools>
For each tool call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:
<tool_call>
{"name": <function-name>, "arguments": <args-json-object>}
</tool_call>. If a tool does not exist in the provided list of tools, notify the user that you do not have the ability to fulfill the request.<|end_of_text|>
<|start_of_role|>user<|end_of_role|>What's the weather like in Boston right now?<|end_of_text|>
<|start_of_role|>assistant<|end_of_role|><tool_call>
{"name": "get_current_weather", "arguments": {"city": "Boston"}}
</tool_call><|end_of_text|>
Evaluation Results:
| Benchmarks | Metric | Micro Dense | H Micro Dense | H Tiny MoE | H Small MoE |
|---|---|---|---|---|---|
| General Tasks | |||||
| MMLU | 5-shot | 65.98 | 67.43 | 68.65 | 78.44 |
| MMLU-Pro | 5-shot, CoT | 44.5 | 43.48 | 44.94 | 55.47 |
| BBH | 3-shot, CoT | 72.48 | 69.36 | 66.34 | 81.62 |
| AGI EVAL | 0-shot, CoT | 64.29 | 59 | 62.15 | 70.63 |
| GPQA | 0-shot, CoT | 30.14 | 32.15 | 32.59 | 40.63 |
| Alignment Tasks | |||||
| AlpacaEval 2.0 | 29.49 | 31.49 | 30.61 | 42.48 | |
| IFEval | Instruct, Strict | 85.5 | 86.94 | 84.78 | 89.87 |
| IFEval | Prompt, Strict | 79.12 | 81.71 | 78.1 | 85.22 |
| IFEval | Average | 82.31 | 84.32 | 81.44 | 87.55 |
| ArenaHard | 25.84 | 36.15 | 35.75 | 46.48 | |
| Math Tasks | |||||
| GSM8K | 8-shot | 85.45 | 81.35 | 84.69 | 87.27 |
| GSM8K Symbolic | 8-shot | 79.82 | 77.5 | 81.1 | 87.38 |
| Minerva Math | 0-shot, CoT | 62.06 | 66.44 | 69.64 | 74 |
| DeepMind Math | 0-shot, CoT | 44.56 | 43.83 | 49.92 | 59.33 |
| Code Tasks | |||||
| HumanEval | pass@1 | 80 | 81 | 83 | 88 |
| HumanEval+ | pass@1 | 72 | 75 | 76 | 83 |
| MBPP | pass@1 | 72 | 73 | 80 | 84 |
| MBPP+ | pass@1 | 64 | 64 | 69 | 71 |
| CRUXEval-O | pass@1 | 41.5 | 41.25 | 39.63 | 50.25 |
| BigCodeBench | pass@1 | 39.21 | 37.9 | 41.06 | 46.23 |
| Tool Calling Tasks | |||||
| BFCL v3 | 59.98 | 57.56 | 57.65 | 64.69 | |
| Multilingual Tasks | |||||
| MULTIPLE | pass@1 | 49.21 | 49.46 | 55.83 | 57.37 |
| MMMLU | 5-shot | 55.14 | 55.19 | 61.87 | 69.69 |
| INCLUDE | 5-shot | 51.62 | 50.51 | 53.12 | 63.97 |
| MGSM | 8-shot | 28.56 | 44.48 | 45.36 | 38.72 |
| Safety | |||||
| SALAD-Bench | 97.06 | 96.28 | 97.77 | 97.3 | |
| AttaQ | 86.05 | 84.44 | 86.61 | 86.64 |
| Benchmarks | # Langs | Languages |
|---|---|---|
| MMMLU | 11 | ar, de, en, es, fr, ja, ko, pt, zh, bn, hi |
| INCLUDE | 14 | hi, bn, ta, te, ar, de, es, fr, it, ja, ko, nl, pt, zh |
| MGSM | 5 | en, es, fr, ja, zh |
Model Architecture:
Granite-4.0-Micro baseline is built on a decoder-only dense transformer architecture. Core components of this architecture are: GQA, RoPE, MLP with SwiGLU, RMSNorm, and shared input/output embeddings.
| Model | Micro Dense | H Micro Dense | H Tiny MoE | H Small MoE |
|---|---|---|---|---|
| Embedding size | 2560 | 2048 | 1536 | 4096 |
| Number of layers | 40 attention | 4 attention / 36 Mamba2 | 4 attention / 36 Mamba2 | 4 attention / 36 Mamba2 |
| Attention head size | 64 | 64 | 128 | 128 |
| Number of attention heads | 40 | 32 | 12 | 32 |
| Number of KV heads | 8 | 8 | 4 | 8 |
| Mamba2 state size | - | 128 | 128 | 128 |
| Number of Mamba2 heads | - | 64 | 48 | 128 |
| MLP / Shared expert hidden size | 8192 | 8192 | 1024 | 1536 |
| Num. Experts | - | - | 64 | 72 |
| Num. active Experts | - | - | 6 | 10 |
| Expert hidden size | - | - | 512 | 768 |
| MLP activation | SwiGLU | SwiGLU | SwiGLU | SwiGLU |
| Sequence length | 128K | 128K | 128K | 128K |
| Position embedding | RoPE | NoPE | NoPE | NoPE |
| # Parameters | 3B | 3B | 7B | 32B |
| # Active parameters | 3B | 3B | 1B | 9B |
Training Data: Overall, our SFT data is largely comprised of three key sources: (1) publicly available datasets with permissive license, (2) internal synthetic data targeting specific capabilities, and (3) a select set of human-curated data.
Infrastructure: We trained the Granite 4.0 Language Models utilizing an NVIDIA GB200 NVL72 cluster hosted in CoreWeave. Intra-rack communication occurs via the 72-GPU NVLink domain, and a non-blocking, full Fat-Tree NDR 400 Gb/s InfiniBand network provides inter-rack communication. This cluster provides a scalable and efficient infrastructure for training our models over thousands of GPUs.
Ethical Considerations and Limitations: Granite 4.0 Instruction Models are primarily finetuned using instruction-response pairs mostly in English, but also multilingual data covering multiple languages. Although this model can handle multilingual dialog use cases, its performance might not be similar to English tasks. In such case, introducing a small number of examples (few-shot) can help the model in generating more accurate outputs. While this model has been aligned by keeping safety in consideration, the model may in some cases produce inaccurate, biased, or unsafe responses to user prompts. So we urge the community to use this model with proper safety testing and tuning tailored for their specific tasks.
Resources