Instructions to use SandLogicTechnologies/Mellum2-12B-A2.5B-Base-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use SandLogicTechnologies/Mellum2-12B-A2.5B-Base-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="SandLogicTechnologies/Mellum2-12B-A2.5B-Base-GGUF",
	filename="Mellum2-12B-A2.5B-Base_IQ3_M.gguf",
)

output = llm(
	"Once upon a time,",
	max_tokens=512,
	echo=True
)
print(output)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use SandLogicTechnologies/Mellum2-12B-A2.5B-Base-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/Mellum2-12B-A2.5B-Base-GGUF:IQ3_M
# Run inference directly in the terminal:
llama-cli -hf SandLogicTechnologies/Mellum2-12B-A2.5B-Base-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/Mellum2-12B-A2.5B-Base-GGUF:IQ3_M
# Run inference directly in the terminal:
llama-cli -hf SandLogicTechnologies/Mellum2-12B-A2.5B-Base-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/Mellum2-12B-A2.5B-Base-GGUF:IQ3_M
# Run inference directly in the terminal:
./llama-cli -hf SandLogicTechnologies/Mellum2-12B-A2.5B-Base-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/Mellum2-12B-A2.5B-Base-GGUF:IQ3_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf SandLogicTechnologies/Mellum2-12B-A2.5B-Base-GGUF:IQ3_M

Use Docker

docker model run hf.co/SandLogicTechnologies/Mellum2-12B-A2.5B-Base-GGUF:IQ3_M

LM Studio
Jan

vLLM

How to use SandLogicTechnologies/Mellum2-12B-A2.5B-Base-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "SandLogicTechnologies/Mellum2-12B-A2.5B-Base-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SandLogicTechnologies/Mellum2-12B-A2.5B-Base-GGUF",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/SandLogicTechnologies/Mellum2-12B-A2.5B-Base-GGUF:IQ3_M

Ollama
How to use SandLogicTechnologies/Mellum2-12B-A2.5B-Base-GGUF with Ollama:
```
ollama run hf.co/SandLogicTechnologies/Mellum2-12B-A2.5B-Base-GGUF:IQ3_M
```

Unsloth Studio

How to use SandLogicTechnologies/Mellum2-12B-A2.5B-Base-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/Mellum2-12B-A2.5B-Base-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/Mellum2-12B-A2.5B-Base-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/Mellum2-12B-A2.5B-Base-GGUF to start chatting

Atomic Chat new
Docker Model Runner
How to use SandLogicTechnologies/Mellum2-12B-A2.5B-Base-GGUF with Docker Model Runner:
```
docker model run hf.co/SandLogicTechnologies/Mellum2-12B-A2.5B-Base-GGUF:IQ3_M
```

Lemonade

How to use SandLogicTechnologies/Mellum2-12B-A2.5B-Base-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull SandLogicTechnologies/Mellum2-12B-A2.5B-Base-GGUF:IQ3_M

Run and chat with the model

lemonade run user.Mellum2-12B-A2.5B-Base-GGUF-IQ3_M

List all available models

lemonade list

Mellum2-12B-A2.5B-Base

Mellum2-12B-A2.5B-Base is a foundation language model developed by JetBrains and optimized for software engineering, code generation, code completion, and code understanding workflows. This repository contains GGUF quantized variants of the model optimized for efficient local inference using llama.cpp.

Built upon the Mellum2 architecture, the model is designed as a pretrained foundation model for software development tasks and serves as the base for downstream instruction-tuned and reasoning-focused variants. The quantized formats significantly reduce memory requirements while preserving strong coding capabilities, enabling practical deployment across consumer hardware and local development environments.

The model is intended primarily for code-centric applications, research, fine-tuning, and software engineering workflows rather than conversational assistant use cases.

Model Overview

Model Name: Mellum2-12B-A2.5B-Base
Base Model: JetBrains/Mellum2-12B-A2.5B-Base
Architecture: Mixture-of-Experts (MoE) Transformer
Parameter Count: 12 Billion Total Parameters / 2.5 Billion Active Parameters
Experts: 64 Experts / 8 Active Experts per Token
Context Window: 131K Tokens
Modalities: Text
Primary Languages: English
Developer: JetBrains
License: Apache 2.0

Quantization Formats

This repository provides various GGUF quantized versions of the Mellum2-12B-A2.5B-Base model, optimized for efficient local inference using llama.cpp. Below are the details of the available I-Matrix (IQ) formats.

IQ3_M

Size reduction of approx 75.71% (5.50 GB) compared to 16-bit (22.64 GB)
Aggressive 3-bit quantization optimized for maximum memory efficiency
Suitable for CPU-only inference and low-memory deployment environments
Enables practical execution of code generation and code completion workloads on constrained hardware
Output quality may reduce on complex repository-level understanding and long-context development tasks

IQ4_NL

Size reduction of approx 71.25% (6.51 GB) compared to 16-bit (22.64 GB)
Advanced 4-bit non-linear quantization designed to better preserve coding quality and code-completion capability
More suitable for software engineering workflows, source-code generation, and technical development tasks
Designed to reduce quantization loss compared to more aggressive formats
Slightly increased computational overhead during inference

IQ4_XS

Size reduction of approx 72.31% (6.27 GB) compared to 16-bit (22.64 GB)
Balanced 4-bit quantization focused on efficiency and stable coding performance
Good trade-off between model size, generation quality, and inference speed
Suitable for code completion, source-code generation, and developer-oriented workflows
Maintains reliable generation behavior across most practical software engineering workloads

Training Background (Original Model)

Mellum2-12B-A2.5B-Base is trained with an emphasis on software engineering, source-code understanding, code generation, and developer productivity workflows.

Pretraining

Large-scale training across programming languages, software repositories, technical documentation, and development resources
Focus on code understanding, code completion, and software engineering tasks
Optimized for downstream coding and technical development workloads

Foundation Model Objectives

Trained as a general-purpose software engineering foundation model
Designed to support downstream fine-tuning and instruction tuning workflows
Optimized for strong source-code representation and generation capabilities
Serves as the foundation for specialized Mellum2 variants

Key Capabilities

Code Generation Supports generation of source code across multiple programming languages and development workflows.
Code Completion Assists with intelligent code completion and developer productivity tasks.
Code Understanding Helps analyze, explain, and understand existing codebases.
Software Engineering Workflows Supports implementation planning, refactoring, and development-oriented tasks.
Efficient Local Deployment Quantized variants enable practical offline inference on consumer hardware.

Usage Example

Using llama.cpp

./llama-cli \
  -m SandlogicTechnologies/Mellum2-12B-A2.5B-Base_IQ4_NL.gguf \
  -p "Write a Python implementation of a binary search tree."

Recommended Usecases

Code Completion Systems Power local IDE integrations and developer productivity tools.
Code Generation Workflows Generate functions, classes, libraries, and application components.
Software Engineering Research Evaluate coding models and software-development benchmarks.
Model Fine-Tuning Serve as a foundation model for downstream instruction-tuning and domain adaptation.
Developer Tooling Build local coding assistants, source-code analysis tools, and development workflows.

Acknowledgments

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

Special thanks to:

The JetBrains team for developing and releasing the Mellum2-12B-A2.5B-Base model.
Georgi Gerganov and the llama.cpp open-source community for enabling efficient quantization and inference via the GGUF format.

Contact

For questions, feedback, or support, please reach out at support@sandlogic.com or visit https://www.sandlogic.com/

Downloads last month: 64

GGUF

Model size

12B params

Architecture

mellum

Hardware compatibility

3-bit

4-bit

Model tree for SandLogicTechnologies/Mellum2-12B-A2.5B-Base-GGUF

Base model

JetBrains/Mellum2-12B-A2.5B-Base

Quantized

(3)

this model