Irfanuruchi/qwen2.5-3b-buildeng

Qwen2.5-3B BuildEng

Repository: Irfanuruchi/qwen2.5-3b-buildeng

Qwen2.5-3B BuildEng is the mid-sized branch of the BuildEng project, based on Qwen/Qwen2.5-3B-Instruct.

BuildEng is my domain-specialized engineering language model project focused on civil engineering, structural reasoning, construction workflows, building diagnostics, and conservative engineering-assistant behavior.

The 3B branch sits between the lightweight 1.5B model and the larger 32B model. Its goal is to improve reasoning consistency, structural routing, construction-stage reasoning, and engineering behavior while still remaining practical for local inference and lighter hardware.

Base Model

Qwen/Qwen2.5-3B-Instruct

Dataset

Dataset used:

Irfanuruchi/buildeng-v8-3b

Training

Training method:

QLoRA fine-tuning with Unsloth.

This repository contains the full merged model weights.

BuildEng Model Family

BuildEng includes multiple branches.

The 1.5B branch is the lightweight model focused on efficient local use and basic conservative engineering behavior.

The 3B branch improves reasoning consistency and civil/building engineering coverage while staying much lighter than the 32B model.

The 32B branch is the flagship large-scale BuildEng model focused on deeper structural reasoning, longer workflows, and stronger inspection-first behavior.

Main Focus

BuildEng 3B focuses on civil and building engineering workflows, including reinforced concrete, beams, slabs, columns, foundations, retaining walls, settlement reasoning, structural connections, temporary works, excavation safety, construction sequencing, waterproofing, durability, and renovation uncertainty.

The model is intended to be cautious with structural questions. It should avoid unsupported approval, separate symptoms from diagnosis, ask for missing information, and recommend inspection or qualified engineering review when safety may be involved.

Important Limitation

This model is not a licensed engineer and must not be used as final engineering approval, design certification, construction sign-off, or replacement for professional review.

It is intended for research, education, drafting support, engineering-assistant workflows, and preliminary reasoning only.

Author

Irfan Uruchi

Part of my ongoing work on domain-specialized engineering language models, structural reasoning datasets, and practical civil/building engineering AI systems.