Irfanuruchi/qwen2.5-1.5b-buildeng

Overview

Irfanuruchi/qwen2.5-1.5b-buildeng is a specialized 1.5 billion parameter language model, fine-tuned from Qwen2.5-1.5B-Instruct using QLoRA with Unsloth. Its primary focus is on civil and building engineering reasoning, aiming to provide more careful and conservative responses to engineering prompts, especially when input data is incomplete or potentially unsafe. The model was trained on the Irfanuruchi/buildeng-v8-1.5b dataset, which includes repair-focused and routing-focused data to address common failure points in earlier engineering models.

Key Capabilities

• Structural Reasoning: Handles tasks related to beams, columns, slabs, retaining walls, foundations, and masonry.
• Construction Decision-Making: Addresses construction sequencing, temporary works, excavation safety, and renovation cases with missing information.
• Conservative Judgment: Specifically trained to reduce "confident wrong answers" by incorporating data around common engineering pitfalls.
• Problem Identification: Can identify issues like incorrect beam/column behavior mixing, generic drainage advice, or approving from insufficient sketches.

Good For

• Engineering-Assistant Workflows: Intended for research and assisting engineers in preliminary analysis or understanding complex scenarios.
• Identifying Risks: Useful for flagging potentially unsafe conditions or incomplete information in engineering contexts.
• Educational Purposes: Can serve as a tool for understanding fundamental civil/building engineering principles and common design considerations.

Limitations

This model is for research and assistant workflows only; final engineering approval must come from a licensed engineer. It uses synthetic and simplified training data, meaning full professional design workflows (e.g., Eurocode, ACI) are not fully implemented. Users should always verify outputs with a qualified engineer, as the model can still make reasoning or arithmetic mistakes.