Granite-4.0-Tiny-Base-Preview

Model Summary:

Granite-4.0-Tiny-Base-Preview is a 7B-parameter hybrid mixture-of-experts (MoE) language model featuring a 128k token context window. The architecture leverages Mamba-2, superimposed with a softmax attention for enhanced expressiveness, with no positional encoding for better length generalization.

• Developers: Granite Team, IBM
• Website: Granite Docs
• Release Date: May 2nd, 2025
• License: Apache 2.0

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 12 languages.

Intended Use: Prominent use cases of LLMs in text-to-text generation include summarization, text classification, extraction, question-answering, and other long-context tasks. All Granite Base models are able to handle these tasks as they were trained on a large amount of data from various domains. Moreover, they can serve as baseline to create specialized models for specific application scenarios.

Installation:

While the native support of this model in Hugging Face Transformers is pending (PR), you need to install transformers from the following source to use this model:

git clone https://github.com/Ssukriti/transformers.git
cd transformers
git checkout granitemoe_hybrid_external_cleanup
pip install -e .

Generation: After installation, copy the code snippet below to run the example.

from transformers import AutoModelForCausalLM, AutoTokenizer
device = "auto"
model_path = "ibm-granite/granite-4.0-tiny-base-preview"
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
input_text = "Where is the Thomas J. Watson Research Center located?"
# tokenize the text
input_tokens = tokenizer(input_text, return_tensors="pt").to(device)
# generate output tokens
output = model.generate(**input_tokens,
                        max_length=4000)
# decode output tokens into text
output = tokenizer.batch_decode(output)
# print output
print(output)

Evaluation Results:

Comparison with 3.3 Base models¹

Models	ARC-Challenge	Hellaswag	MMLU	TruthfulQA	Winogrande	GSM8K	DROP	NQ	AGIEval	TriviaQA	Avg
Granite-3.3-2B-Base	47.49	73.2	54.33	40.83	70.4	50.0	32.552	24.36	38.78	63.22	49.52
Granite-3.3-8B-Base	50.84	80.1	63.89	52.15	74.4	59.0	36.14	36.5	49.3	78.18	58.05
Granite-4.0-Tiny-Base-Preview	54.52	75.80	57.86	44.57	71.1	49.0	41.74	28.48	42.61	67.85	53.35

Model Architecture: Granite-4.0-Tiny-Base-Preview is based on a decoder-only dense transformer architecture. Core components of this architecture are: GQA and NoPE, MLP with SwiGLU, RMSNorm, and shared input/output embeddings.

Model	2B Dense	8B Dense	Granite-4.0-Tiny-Base-Preview
Embedding size	2048	4096	1536
Number of layers	40	40	40
Attention head size	64	128	128
Number of attention heads	32	32	12
Number of KV heads	8	8	4
MLP hidden size	8192	12800	512
MLP activation	SwiGLU	SwiGLU	SwiGLU
Initialization std	0.1	0.1	0.1
Sequence length	128K	128K	128K
Position embedding	RoPE	RoPE	None
# Parameters	2.5B	8.1B	6.7B
# Active parameters	2.5B	8.1B	1B
# Training tokens	12T	12T	2.5T

Training Data: This model is trained on a mix of open source and proprietary data following a two-stage training strategy.

• Stage 1 data: The data for stage 1 is sourced from diverse domains, such as: web, code, academic sources, books, and math data.
• Stage 2 data: The data for stage 2 comprises a curated mix of high-quality data from the same domains, plus multilingual and instruction data. The goal of this second training phase is to enhance the model’s performance on specific tasks. contains a recitation of the related paragraph before the answer.

Infrastructure: We train Granite 4.0 Language Models using IBM's super computing cluster, Blue Vela, which is outfitted with NVIDIA H100 GPUs. This cluster provides a scalable and efficient infrastructure for training our models over thousands of GPUs.

Ethical Considerations and Limitations: The use of Large Language Models involves risks and ethical considerations people must be aware of, including but not limited to: bias and fairness, misinformation, and autonomous decision-making. Granite-4.0-Tiny-Base-Preview model is not the exception in this regard. Even though this model is suited for multiple generative AI tasks, it has not undergone any safety alignment, there it may produce problematic outputs. Additionally, it remains uncertain whether smaller models might exhibit increased susceptibility to hallucination in generation scenarios by copying text verbatim from the training dataset due to their reduced sizes and memorization capacities. This aspect is currently an active area of research, and we anticipate more rigorous exploration, comprehension, and mitigations in this domain. Regarding ethics, a latent risk associated with all Large Language Models is their malicious utilization. We urge the community to use Granite-4.0-Tiny-Base-Preview model with ethical intentions and in a responsible way.

Resources

• ⭐️ Learn about the latest updates with Granite: https://www.ibm.com/granite
• 📄 Get started with tutorials, best practices, and prompt engineering advice: https://www.ibm.com/granite/docs/
• 💡 Learn about the latest Granite learning resources: https://ibm.biz/granite-learning-resources

[1] Evaluated using OLMES