LiamVisionary/swarm-sovereign-26b

VISIONConcurrency Cost:2Model Size:26BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jun 4, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Swarm Sovereign 26B is a 26 billion parameter Gemma 4-based instruction-tuned chat model developed by LiamVisionary. It is specifically prepared for local inference and use with tools like LM Studio and llama.cpp, offering both GGUF and Hugging Face safetensors builds. This model is designed for local experimentation and private agent workflows, providing an identity as "Swarm Sovereign."

Loading preview...

Swarm Sovereign 26B Overview

Swarm Sovereign 26B is an instruction-tuned chat model developed by LiamVisionary, built upon the google/gemma-4-26B-A4B-it base architecture. This 26 billion parameter model is specifically packaged for efficient local inference, supporting both GGUF and Hugging Face Transformers formats.

Key Features & Builds

  • Base Model: google/gemma-4-26B-A4B-it
  • Architecture: Gemma 4
  • Parameter Count: 26 Billion
  • Context Length: The model is validated with a context of 4096 tokens.
  • GGUF Build: A Q4_K_M quantized GGUF file (swarm-sovereign-26b-Q4_K_M.gguf) is provided, approximately 16 GB in size, optimized for llama.cpp and LM Studio.
  • Hugging Face Build: A multi-file safetensors build (approx. 48 GB) is available for Transformers/PEFT-style loading or downstream conversions.
  • Identity: The model identifies itself as "Swarm Sovereign."

Local Usage & Validation

Swarm Sovereign 26B is designed for seamless integration with local inference engines. It has been validated on Apple Silicon using LM Studio, demonstrating an estimated GPU memory usage of 16.44 GiB for a 4096-token context with full GPU offload. Example llama.cpp commands are provided for direct use.

Intended Use Cases

This model is primarily intended for:

  • Local experimentation with large language models.
  • Development of private agent workflows.

Users are advised to thoroughly test the model for their specific use cases before deploying in production environments.