Granite-3.1-2B-Reasoning-LORA (Efficient Fine-Tuned Model)

Model Overview

This model is a LoRA fine-tuned version of ibm-granite/granite-3.1-2b-instruct, optimized for reasoning tasks while maintaining efficiency and low computational cost. Using LoRA (Low-Rank Adaptation), this model retains the full power of the base model while applying targeted modifications for logical and analytical reasoning.

• Developed by: ruslanmv
• License: Apache 2.0
• Base Model: ibm-granite/granite-3.1-2b-instruct
• Fine-tuned for: Logical reasoning, structured problem-solving, long-context tasks
• Training Method: LoRA (Low-Rank Adaptation)
• Supported Languages: English

---

Why Use the LoRA Version?

This LoRA fine-tuned model provides several benefits:

✅ Memory-efficient fine-tuning with LoRA
✅ 2x Faster Training using Unsloth and Hugging Face TRL
✅ Retains the base model’s capabilities while enhancing reasoning skills
✅ Easier to merge with other adapters or apply to specific tasks

---

Installation & Usage

To use this LoRA fine-tuned model, install the necessary dependencies:

pip install torch torchvision torchaudio
pip install accelerate
pip install transformers
pip install peft
pip install bitsandbytes

Running the Model

Load and merge the LoRA adapter with the base model:

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch

device = "cuda" if torch.cuda.is_available() else "cpu"
base_model_path = "ibm-granite/granite-3.1-2b-instruct"
lora_model_path = "ruslanmv/granite-3.1-2b-Reasoning-LORA"

tokenizer = AutoTokenizer.from_pretrained(base_model_path)
model = AutoModelForCausalLM.from_pretrained(base_model_path, device_map="auto")

# Load LoRA adapter
model = PeftModel.from_pretrained(model, lora_model_path)
model.eval()

input_text = "Can you explain the difference between inductive and deductive reasoning?"
input_tokens = tokenizer(input_text, return_tensors="pt").to(device)

output = model.generate(**input_tokens, max_length=4000)
output_text = tokenizer.batch_decode(output)

print(output_text)

---

Intended Use

Granite-3.1-2B-Reasoning-LORA is optimized for efficient reasoning while keeping computational costs low, making it ideal for:

• Logical and analytical problem-solving
• Text-based reasoning tasks
• Mathematical and symbolic reasoning
• Advanced instruction-following

This LoRA-based fine-tuning method is particularly useful for lightweight deployment and quick adaptability to specific tasks.

---

License & Acknowledgments

This model is released under the Apache 2.0 license. It is fine-tuned from IBM’s Granite 3.1-2B-Instruct model using LoRA fine-tuning. Special thanks to the IBM Granite Team for developing the base model.

For more details, visit the IBM Granite Documentation.

---

Citation

If you use this model in your research or applications, please cite:

@misc{ruslanmv2025granite,
  title={LoRA Fine-Tuning of Granite-3.1 for Advanced Reasoning},
  author={Ruslan M.V.},
  year={2025},
  url={https://huggingface.co/ruslanmv/granite-3.1-2b-Reasoning-LORA}
}