Model Overview

The arungovindneelan/foam-cfd-unified-7b is a specialized 7.6 billion parameter language model, fine-tuned from the Qwen2.5-Coder-7B-Instruct base model. Its core purpose is to automate and streamline Computational Fluid Dynamics (CFD) workflows, specifically focusing on generating necessary configuration files from natural language descriptions.

Key Capabilities

• Mesh Generation: Translates plain-English geometry descriptions into executable Gmsh .geo scripts for mesh creation.
• OpenFOAM Case File Generation: Produces OpenFOAM boundary condition files (e.g., 0/U, 0/p) based on flow prompts and mesh patch information.
• BC Patch Specification: Generates compact JSON diffs for boundary condition patches, allowing for easy modification of existing templates.
• End-to-End Simulation: Designed to integrate with a server for a full simulation pipeline, from prompt to running OpenFOAM cases and returning results.

Training and Performance

The model was fine-tuned using QLoRA (4-bit NF4) on a unified dataset of 20,505 examples, covering mesh generation, OpenFOAM file generation, and various BC patch specifications. It requires approximately 6 GB of VRAM for 4-bit inference and supports a sequence length of 4096 tokens. The model is intended for use on Linux systems with Python 3.10+, an NVIDIA GPU, and CUDA 12.x, with OpenFOAM 11 required for full simulation capabilities.