2rmp-agent** is a specialized, highly-efficient 1.5B parameter language model fine-tuned specifically for route optimization and geospatial data processing. It is designed to act as the “brain” for the v2rmp (Vehicle Routing Problem & Map Processing) ecosystem, allowing for 100% offline, air-gapped routing orchestration. 4 5 This model is a QLoRA fine-tune of Qwen/Qwen2.5-1.5B-Instruct and has been quantized to 4-bit (Q4_K_M) for blazing-fast inference on almost any hardware (including edge devices and laptops without dedicated GPUs). 6 7 ## 🚀 Capabilities 8 9 When connected to the v2rmp core, this agent can autonomously handle: 10 - Map Extraction: Pulling road networks from OpenStreetMap (OSM) PBF files, PMTiles, PostGIS, or Overture Maps. 11 - Graph Compilation: Cleaning GeoJSON and compiling highly-compressed .rmp binary graphs. 12 - VRP & CPP Optimization: Orchestrating complex Vehicle Routing Problems (VRP) and Chinese Postman Problems (CPP) using Clarke-Wright, Sweep, Neural GNN, and Local Search solvers. 13 - Elevation & Fuel: Querying DEM GeoTIFFs for elevation profiles and drone fuel consumption estimates. 14 15 ## 💻 Usage 16 17 To run the model locally via Ollama: ollama run spacialglaciercom/v2rmp-agent

1
2 ### Example Prompts
3 Try asking the agent routing-specific questions:
4
5 * *"How do I extract a road network from a local OSM PBF file for Montreal?"*
6 * *"Optimize a route using the Clarke-Wright solver for 5 vehicles with a capacity of 1000, starting at depot
  45.505, -73.565."*
7 * *"What's the command to compile my raw GeoJSON into a clean .rmp binary map and prune disconnected
  subgraphs?"*
8
9 ## 🔒 100% Offline Orchestration

10 By combining this Ollama model with locally downloaded .osm.pbf map files and the v2rmp Rust binaries, you can achieve a completely air-gapped route optimization pipeline without relying on external APIs like Google Maps or Valhalla. 11 12 ## Model Details 13 * Base Model: Qwen/Qwen2.5-1.5B-Instruct 14 * Parameters: 1.5B 15 * Quantization: Q4_K_M (4-bit) 16 * Training Method: Supervised Fine-Tuning (SFT) / QLoRA 17 * License: Apache 2.0 (Inherited)