Instructions to use SandLogicTechnologies/Fara1.5-4B-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use SandLogicTechnologies/Fara1.5-4B-GGUF with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf SandLogicTechnologies/Fara1.5-4B-GGUF:IQ3_M
# Run inference directly in the terminal:
llama cli -hf SandLogicTechnologies/Fara1.5-4B-GGUF:IQ3_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf SandLogicTechnologies/Fara1.5-4B-GGUF:IQ3_M
# Run inference directly in the terminal:
llama cli -hf SandLogicTechnologies/Fara1.5-4B-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/Fara1.5-4B-GGUF:IQ3_M
# Run inference directly in the terminal:
./llama-cli -hf SandLogicTechnologies/Fara1.5-4B-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/Fara1.5-4B-GGUF:IQ3_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf SandLogicTechnologies/Fara1.5-4B-GGUF:IQ3_M

Use Docker

docker model run hf.co/SandLogicTechnologies/Fara1.5-4B-GGUF:IQ3_M

LM Studio
Jan

vLLM

How to use SandLogicTechnologies/Fara1.5-4B-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "SandLogicTechnologies/Fara1.5-4B-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/Fara1.5-4B-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/SandLogicTechnologies/Fara1.5-4B-GGUF:IQ3_M

Ollama
How to use SandLogicTechnologies/Fara1.5-4B-GGUF with Ollama:
```
ollama run hf.co/SandLogicTechnologies/Fara1.5-4B-GGUF:IQ3_M
```
Unsloth Desktop

How to use SandLogicTechnologies/Fara1.5-4B-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf SandLogicTechnologies/Fara1.5-4B-GGUF:IQ3_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "SandLogicTechnologies/Fara1.5-4B-GGUF:IQ3_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use SandLogicTechnologies/Fara1.5-4B-GGUF with Docker Model Runner:
```
docker model run hf.co/SandLogicTechnologies/Fara1.5-4B-GGUF:IQ3_M
```

Lemonade

How to use SandLogicTechnologies/Fara1.5-4B-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull SandLogicTechnologies/Fara1.5-4B-GGUF:IQ3_M

Run and chat with the model

lemonade run user.Fara1.5-4B-GGUF-IQ3_M

List all available models

lemonade list

Hermes Agent

How to use SandLogicTechnologies/Fara1.5-4B-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf SandLogicTechnologies/Fara1.5-4B-GGUF:IQ3_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default SandLogicTechnologies/Fara1.5-4B-GGUF:IQ3_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use SandLogicTechnologies/Fara1.5-4B-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf SandLogicTechnologies/Fara1.5-4B-GGUF:IQ3_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "SandLogicTechnologies/Fara1.5-4B-GGUF:IQ3_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

FrogNano-4B-2609

FrogNano-4B-2609 is a compact repository-level coding agent developed by Microsoft and derived from Qwen3.5-4B. It is specifically post-trained for software-engineering tasks, with a focus on repository navigation, debugging, code editing, test execution, and candidate patch generation.

The model is trained using reinforcement learning on approximately 1,500 synthetic software-engineering tasks generated and calibrated through TaskPilot. FrogNano operates through the lightweight five-tool Leaf harness, allowing it to inspect and search repositories, modify files, execute shell commands, and run tests as part of an iterative coding workflow.

The GGUF variants in this repository provide reduced-size representations of FrogNano-4B-2609 for local inference while maintaining the model's compact coding-agent architecture.

Model Overview

Model Name: FrogNano-4B-2609
Base Model: Qwen/Qwen3.5-4B
Parameter Count: Approximately 4.66 Billion Parameters
Context Length: Approximately 131K tokens
Modalities: Text
Primary Language: English
Developer: Microsoft
License: Apache 2.0

FrogNano is specifically intended and evaluated for text-based repository-level software engineering. Although the upstream Qwen3.5-4B architecture contains image and video components, these modalities were not post-trained or evaluated for FrogNano and are not claimed as supported.

Quantization Formats

This repository provides multiple GGUF quantized versions of FrogNano-4B-2609 optimized for reduced storage and memory requirements. The available formats provide different trade-offs between model footprint and numerical precision.

IQ4_NL

Size reduction of approx. 69.02% (2.50 GB) compared to 16-bit (8.07 GB)
4-bit non-linear quantization designed to provide a compact representation of the model weights.
Provides a substantial reduction in memory requirements while retaining more weight information than the supplied 3-bit variant.
Suitable for local coding-agent workloads where reduced model size is important.
Provides a practical option for repository analysis, debugging, code generation, and software-engineering tasks.

IQ4_XS

Size reduction of approx. 70.01% (2.42 GB) compared to 16-bit (8.07 GB)
Compact 4-bit quantization intended for deployments with tighter storage and memory constraints.
Offers a smaller file footprint than IQ4_NL while retaining a 4-bit weight representation.
Suitable for local code-generation and repository-analysis workflows.
Provides an efficient option when minimizing model storage is a higher priority.

IQ3_M

Size reduction of approx. 74.47% (2.06 GB) compared to 16-bit (8.07 GB)
Aggressive 3-bit quantization that provides the smallest GGUF representation among the supplied variants.
Reduces memory and storage requirements significantly for constrained local environments.
Suitable for lightweight coding-agent experimentation and repository-level analysis where model footprint is a primary consideration.
The lower numerical precision can introduce a greater trade-off for demanding reasoning and complex software-engineering tasks.

Q6_K

Size reduction of approx. 58.86% (3.32 GB) compared to 16-bit (8.07 GB)
Higher-precision 6-bit K-Quant representation designed to retain more numerical information than the lower-bit variants.
Provides the largest file size among the supplied quantized versions while remaining substantially smaller than the F16 model.
Suitable for coding workloads where preserving a higher-precision weight representation is preferred over maximum compression.
Can be considered for more demanding repository reasoning, debugging, and code-generation workloads.

Training Background (Original Model)

FrogNano is derived from Qwen/Qwen3.5-4B and receives additional reinforcement-learning-based post-training specifically focused on repository-level software engineering.

Reinforcement Learning

Approximately 1,500 synthetic software-engineering task environments were generated, validated, and calibrated using TaskPilot.
Training uses complete multi-turn coding trajectories rather than stronger-model solution trajectories.
Executable rewards are used to evaluate functional correctness, tool usage, regression avoidance, and repository-level task completion.
The Leaf harness provides the model with five tools for repository interaction, including file inspection and search, editing, shell execution, and test execution.
The training process focuses on long-horizon software-engineering behavior such as debugging, code modification, test execution, and patch generation.

The model's evaluated coding-agent configuration supports approximately 131K tokens of combined interaction context, with up to 8,192 generated tokens per assistant turn.

Key Capabilities

Repository-Level Software Engineering
Works with existing repositories to analyze issues and generate candidate code changes.
Codebase Navigation
Supports cross-file code understanding, file inspection, search, and identification of relevant implementation areas.
Debugging and Bug Fixing
Designed for diagnosing software issues, modifying affected code, and iterating based on test results.
Multi-File Code Editing
Can generate structured tool calls for making coordinated changes across repository files.
Test-Assisted Development
Uses shell commands and test execution as part of iterative coding workflows.
Long-Context Code Reasoning
Supports approximately 131K tokens of combined interaction context in the evaluated coding-agent configuration.

Usage Example

Using llama.cpp

./llama-cli \
  -m SandlogicTechnologies/FrogNano-4B-2609-IQ4_NL.gguf \
  -p "Inspect this repository issue and explain which files are likely responsible for the problem, what changes should be made, and how the fix should be tested."

Recommended Usecases

Repository-Level Coding Agents
Build research and development workflows for automated software-engineering tasks.
Bug Diagnosis and Repair
Analyze repository issues and generate candidate fixes.
Code Maintenance and Refactoring
Assist with modifying existing code across multiple files.
Test-Driven Development
Combine code changes with shell commands and executable tests for iterative development.
Coding-Agent Research
Experiment with long-horizon repository-level coding and tool-use workflows.
Local Coding Assistance
Deploy compact GGUF variants for local experimentation and development workflows.

FrogNano is best suited to English-language, Python-heavy repositories with reproducible environments, clearly defined tasks, and executable test suites. It should not be treated as a general-purpose assistant or as a source of guaranteed correct or production-ready code.

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 FrogNano-4B-2609 and the associated repository-level coding-agent research.
The Qwen team for developing the Qwen3.5-4B base model from which FrogNano is derived.
Georgi Gerganov and the llama.cpp open-source community for enabling efficient local inference and GGUF-based model deployment.

Contact

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

Downloads last month: 1,282

GGUF

Model size

4B params

Architecture

qwen35

Hardware compatibility

3-bit

4-bit

6-bit

Model tree for SandLogicTechnologies/Fara1.5-4B-GGUF

Base model

microsoft/FrogNano-4B-2609

Quantized

(22)

this model