ResiNetResearchGroup/ResiNet-LLM-topology

ResiNet-LLM-topology

This repository contains the implementation for the submission:

An LLM-Based Framework for Intent-Driven Network Topology Design

Purpose

ResiNet-LLM-topology provides the evaluation pipeline used in our work for comparing generated network topologies against reference designs.

It includes:

• —node-edge level F1 evaluation
• —resilience evaluation
• —scenario data and reference topologies used for evaluation

Folder layout

requirements.txt
README.md
data/
  ├─ results/
  ├─ scenarios/
  ├─ topology_scenario_1.json
  ├─ topology_scenario_2.json
  ├─ topology_scenario_3.json
  └─ topology_scenario_4.json

f1_nodes_functions.py
f1_nodes_main.py
f1_edges_functions.py
f1_edges_main.py
resinet_llm_prompt.py
evaluate_connectivity.py
draw_topology.py
models_configuration.py
resinet_llm_prompt.py
topology_generation.py

Installation

Install the dependencies with:

pip install -r requirements.txt

Quick start

Node F1 evaluation

python f1_nodes_main.py --gen <generated-json-or-folder> --ref <reference.json>

Edge F1 evaluation

python f1_edges_main.py --gen <generated-json-or-folder> --ref <reference.json>

Resilience evaluation

python evaluate_connectivity.py <tested-generated-folder>

Topologies generation

• —Start Ollama server (after installation) :

ollama serve

• —Pull a model you plan to use (models are defined in models_configuration.py)

ollama pull qwen3:32b

• —The code expects the Ollama HTTP API to be reachable at http://localhost:11434 (default).

• —Notes for OpenAI / GPT users: When using OpenAI (GPT) backends, set the OPENAI_API_KEY environment variable before running scripts.

• —Commande:

python topology_generation.py --model_topo <model_key> --use_case 1 --type_ablation <type_ablation>

• —--model_topo: model key from models_configuration.py (e.g. qwen3:32b, mistral-small:24b, or GPT keys when using OpenAI).
• —--use_case: integer selecting the user requirements prompt (see data/scenarios/scenarios.py).
• —--type_ablation: ablation flag for experimental runs.

The scripts produce detailed evaluation logs and summary .txt outputs in the local results/ path next to the evaluated files.

File descriptions

• —f1_nodes_functions.py — helper functions for node normalization, mapping, and F1 calculation.
• —f1_nodes_main.py — driver script for node-based evaluation and overall summary generation.
• —f1_edges_functions.py — helper functions for edge cleanup, structural remapping, and edge F1 scoring.
• —f1_edges_main.py — driver script for edge-based evaluation and final metrics.
• —resinet_llm_prompt.py — prompt utilities for intent-driven topology generation.
• —evaluate_connectivity.py — connectivity and resilience measurement helpers.
• —draw_topology.py — graph drawing utilities.
• —requirements.txt — Python package dependencies.
• —topology generation.py— intent-driven topology generation

Repository usage

This package is designed to let users:

1. 1.inspect the evaluation logic used in the paper,
2. 2.reproduce node and edge F1 scoring on generated topologies with LLMs using the previous prompts,
3. 3.compare generated outputs with reference topologies,

Notes for publication

• —This repository is intended as a reproducibility artifact
• —Updates and releases: The repository may be updated for bug fixes, documentation improvements, or additional examples.