aerialblancaservices/v2rmp-routing-ml

v2rmp Routing ML Models

A collection of lightweight neural models for route optimization (VRP/CPP), built with Candle (pure Rust ML) and trained on synthetic + real-world VRP instances.

Part of the v2rmp project — a Rust TUI/CLI for road network extraction, compilation, and multi-vehicle route optimization.

Models

Model	File	Architecture	Purpose
AutoML Predictor	automl_v2.safetensors	28 → 64 → 5 MLP	Predict instance-aware solver hyperparameters (max iterations, temperature, tabu tenure, cooling rate, neighbourhood radius)
Solver Selector	solver_selector_v2.safetensors	28 → 128 → 64 → 6 MLP	Classify VRP instances to the best algorithm among 6 solvers (default, Clarke-Wright, sweep, Or-Opt, 2-Opt, neural-guided)
Quality Predictor	quality_predictor_v2.safetensors	28 → 64 → 32 → 2 MLP	Predict gap-to-optimal (%) and tour length (km) before solving
Move Scorer	move_scorer_v2.safetensors	16 → 32 → 16 → 1 MLP	Score candidate 2-Opt / Or-Opt moves for neural-guided local search
Graph Embedder	graph_embed.safetensors	2-layer GraphSAGE (10 → 64 → 64)	Produce 64-dim learned embeddings for road network edges

Input Features

All MLP models share the same 28-dim normalized instance feature vector derived from VRP instance statistics (stop count, vehicle count, bounding box spread, demand statistics, distance matrix stats, etc.).

Model Loading (Rust / Candle)

use candle_core::{Device, DType};
use candle_nn::{linear, Linear, Module, VarBuilder};

// Load safetensors
let tensors = candle_core::safetensors::load("solver_selector_v2.safetensors", &device)?;
let vb = VarBuilder::from_tensors(tensors, DType::F32, &device);

// Build layers
let lin1 = linear(28, 128, vb.pp("lin1"))?;
let lin2 = linear(128, 64, vb.pp("lin2"))?;
let lin3 = linear(64, 6, vb.pp("lin3"))?;

See the v2rmp source for full loading code.

License

MIT OR Apache-2.0 (same as v2rmp crate).

Generated by ML Intern

This model repository was generated by ML Intern, an agent for machine learning research and development on the Hugging Face Hub.

• Try ML Intern: https://smolagents-ml-intern.hf.space
• Source code: https://github.com/huggingface/ml-intern

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = 'aerialblancaservices/v2rmp-routing-ml'
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

For non-causal architectures, replace AutoModelForCausalLM with the appropriate AutoModel class.