Instructions to use SandLogicTechnologies/Phi-4-reasoning-plus-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use SandLogicTechnologies/Phi-4-reasoning-plus-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="SandLogicTechnologies/Phi-4-reasoning-plus-GGUF",
	filename="Phi-4-reasoning-plus_IQ3_M.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use SandLogicTechnologies/Phi-4-reasoning-plus-GGUF with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf SandLogicTechnologies/Phi-4-reasoning-plus-GGUF:IQ3_M
# Run inference directly in the terminal:
llama-cli -hf SandLogicTechnologies/Phi-4-reasoning-plus-GGUF:IQ3_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf SandLogicTechnologies/Phi-4-reasoning-plus-GGUF:IQ3_M
# Run inference directly in the terminal:
llama-cli -hf SandLogicTechnologies/Phi-4-reasoning-plus-GGUF:IQ3_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf SandLogicTechnologies/Phi-4-reasoning-plus-GGUF:IQ3_M
# Run inference directly in the terminal:
./llama-cli -hf SandLogicTechnologies/Phi-4-reasoning-plus-GGUF:IQ3_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf SandLogicTechnologies/Phi-4-reasoning-plus-GGUF:IQ3_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf SandLogicTechnologies/Phi-4-reasoning-plus-GGUF:IQ3_M

Use Docker

docker model run hf.co/SandLogicTechnologies/Phi-4-reasoning-plus-GGUF:IQ3_M

LM Studio
Jan

vLLM

How to use SandLogicTechnologies/Phi-4-reasoning-plus-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "SandLogicTechnologies/Phi-4-reasoning-plus-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SandLogicTechnologies/Phi-4-reasoning-plus-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/SandLogicTechnologies/Phi-4-reasoning-plus-GGUF:IQ3_M

Ollama
How to use SandLogicTechnologies/Phi-4-reasoning-plus-GGUF with Ollama:
```
ollama run hf.co/SandLogicTechnologies/Phi-4-reasoning-plus-GGUF:IQ3_M
```

Unsloth Studio

How to use SandLogicTechnologies/Phi-4-reasoning-plus-GGUF 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 SandLogicTechnologies/Phi-4-reasoning-plus-GGUF 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 SandLogicTechnologies/Phi-4-reasoning-plus-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for SandLogicTechnologies/Phi-4-reasoning-plus-GGUF to start chatting

Docker Model Runner
How to use SandLogicTechnologies/Phi-4-reasoning-plus-GGUF with Docker Model Runner:
```
docker model run hf.co/SandLogicTechnologies/Phi-4-reasoning-plus-GGUF:IQ3_M
```

Lemonade

How to use SandLogicTechnologies/Phi-4-reasoning-plus-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull SandLogicTechnologies/Phi-4-reasoning-plus-GGUF:IQ3_M

Run and chat with the model

lemonade run user.Phi-4-reasoning-plus-GGUF-IQ3_M

List all available models

lemonade list

Phi-4-reasoning-plus

Phi-4-reasoning-plus is an advanced reasoning-focused language model developed by Microsoft, optimized for multi-step logical reasoning, mathematical problem solving, scientific analysis, and coding-oriented workflows. The model is fine-tuned from Phi-4 using supervised reasoning traces and reinforcement learning to improve reasoning depth and response quality.

The model is designed for reasoning-intensive applications that require structured thinking, extended context handling, and detailed step-by-step generation. It performs strongly across tasks involving mathematics, algorithmic reasoning, coding, and analytical problem solving while remaining practical for local deployment through quantized GGUF formats.

Phi-4-reasoning-plus is particularly effective for workflows requiring deep reasoning chains, structured outputs, technical explanations, and extended-context conversational reasoning.

Model Overview

Model Name: Phi-4-reasoning-plus
Base Model: microsoft/Phi-4-reasoning-plus
Architecture: Decoder-only Transformer
Parameter Count: 14 Billion parameters
Context Window: Up to 64K tokens
Modalities: Text
Primary Languages: English
Developer: Microsoft
License: MIT

Quantization Details

This repository provides various GGUF quantized versions of the Phi-4-reasoning-plus model, optimized for efficient local inference using llama.cpp. Below are the details of the available I-Matrix (IQ) and K-Quant formats.

Quantization Formats

IQ3_M

Size reduction of approx 76.42% (6.44 GB) compared to 16-bit (27.31 GB)
Aggressive 3-bit quantization optimized for maximum memory reduction and lightweight deployment
Suitable for low-memory environments and CPU-based inference workflows
Enables practical deployment of large reasoning-focused models on constrained hardware
May reduce precision on complex mathematical reasoning and long chain-of-thought tasks

IQ4_XS

Size reduction of approx 71.84% (7.69 GB) compared to 16-bit (27.31 GB)
Balanced 4-bit quantization focused on efficient inference and stable generation quality
Provides a strong trade-off between memory usage, reasoning capability, and response consistency
Suitable for conversational reasoning, coding assistance, and structured analytical tasks
Maintains reliable inference performance across most practical workloads

IQ4_NL

Size reduction of approx 70.56% (8.04 GB) compared to 16-bit (27.31 GB)
Advanced 4-bit non-linear quantization designed to better preserve reasoning quality and structured outputs
More suitable for technical explanations, coding workflows, and multi-step analytical reasoning
Typically provides improved consistency compared to lower-bit formats
Slightly higher computational overhead during inference

Q6_K

Size reduction of approx 58.99% (11.20 GB) compared to 16-bit (27.31 GB)
Higher-precision 6-bit K-Quant format optimized for improved output fidelity and reasoning stability
Preserves more of the original model capability compared to lower-bit quantization formats
Well-suited for demanding reasoning, coding, and scientific analysis workloads
Requires more memory but generally provides stronger consistency and generation quality

Training Overview

Pretraining

The model builds upon the Phi-4 architecture and is trained using a combination of high-quality synthetic data, filtered public-domain sources, and curated reasoning-focused datasets. Training objectives include:

Large-scale language modeling
Mathematical and scientific reasoning
Coding and algorithmic problem solving
Structured chain-of-thought generation

Alignment and Optimization

Post-training refinement focuses on reasoning-intensive instruction following and reinforcement learning for improved reasoning depth.

Optimization objectives include:

Supervised fine-tuning using reasoning traces
Reinforcement learning for reasoning quality improvement
Enhanced multi-step analytical reasoning
Improved long-context coherence and structured generation

Core Capabilities

Advanced Reasoning Handles multi-step logical, mathematical, and analytical tasks with strong structured reasoning ability.
Mathematics and Scientific Problem Solving Performs effectively on reasoning-intensive STEM-oriented workloads.
Coding and Algorithmic Workflows Supports code generation, debugging, and technical explanation tasks.
Long-Context Understanding Maintains coherence and logical consistency across extended context windows.
Instruction Following Produces structured and context-aware responses for complex prompts and reasoning tasks.

Example Usage

llama.cpp

./llama-cli \
  -m SandlogicTechnologies/Phi-4-reasoning-plus_IQ4_NL.gguf \
  -p "Solve this step-by-step: If a train travels 240 km in 3 hours, what is its average speed?"

Recommended Use Cases

Mathematical and scientific reasoning systems
Coding assistants and algorithmic problem-solving workflows
Educational and tutoring applications
Long-context analytical and technical conversations
Structured reasoning and chain-of-thought generation
Research and experimentation involving reasoning-focused LLMs

Acknowledgments

These quantized models are based on the original work by the Microsoft development team.

Special thanks to:

The Microsoft team for developing and releasing the Phi-4-reasoning-plus model.
Georgi Gerganov and the llama.cpp open-source community for enabling efficient quantization and inference via the GGUF format.

Contact

For any inquiries or support, please contact us at support@sandlogic.com or visit our Website.

Downloads last month: 512

GGUF

Model size

15B params

Architecture

phi3

Hardware compatibility

3-bit

4-bit

6-bit

Model tree for SandLogicTechnologies/Phi-4-reasoning-plus-GGUF

Base model

microsoft/phi-4

Finetuned

microsoft/Phi-4-reasoning-plus

Quantized

(44)

this model