Nexus Swarm Worker — GGUF (Qwen3-8B Q4_K_M)

Quantized GGUF of the worker node in the NEXUS Swarm v4a-max distributed offensive security agent system. The full LoRA adapter chain (DAPT + SFT + FARCA-GRPO) is merged into Qwen/Qwen3-8B and quantized to Q4_K_M via llama.cpp.

Role: Worker

The worker is the executor of the swarm. It receives specific, scoped subtasks from the orchestrator and executes them immediately with minimal reasoning. Designed to run on an RTX 3070 Laptop (8 GB).

Core behavior:

• Execute delegated tasks with correct tool calls — no re-planning
• Parse and structure tool output for the orchestrator to consume
• Prioritize speed and correctness over verbose explanation

Files

Pipeline Position

DAPT → SFT → FARCA-GRPO → merge + quantize → GGUF (this model)

Swarm Topology

Quick Start

llama-server

llama-server \
    -m Qwen3-8B-Q4_K_M.gguf \
    --host 0.0.0.0 --port 8084 \
    --ctx-size 4096 \
    -ngl 99

Ollama

# Create a Modelfile
cat > Modelfile <<EOF
FROM ./Qwen3-8B-Q4_K_M.gguf
PARAMETER num_ctx 4096
SYSTEM "You are a worker node. Execute the delegated task immediately with the correct tool call. Keep reasoning to one sentence maximum."
EOF

ollama create nexus-worker -f Modelfile
ollama run nexus-worker

LM Studio

Download Qwen3-8B-Q4_K_M.gguf and load it in LM Studio. Set context length to 4096.

Inference Specifications

• Target hardware: RTX 3070 Laptop (8 GB VRAM)
• Quantization: Q4_K_M
• Context length: 4096 tokens
• GPU layers: All on GPU (-ngl 99)
• Inference port: 8084 (swarm default)

Developed By

Atlas Institute — Code-Trainer (RTPI) pipeline.