arungovindneelan/foam-raft-patch-gen
arungovindneelan/foam-raft-patch-gen is a 7.6 billion parameter Qwen2.5-Coder-7B model fine-tuned by arungovindneelan using Retrieval-Augmented Fine-Tuning (RAFT). It specializes in generating OpenFOAM 11 boundary condition (BC) files via a template-patch approach. The model takes a retrieved tutorial file and a natural-language simulation description to output a JSON patch specification, adapting template BCs for new cases, and includes chain-of-thought reasoning.
Loading preview...
foam-raft-patch-gen: OpenFOAM Boundary Condition Generation
This model, developed by arungovindneelan, is a Qwen2.5-Coder-7B variant with approximately 5.8 billion parameters, fine-tuned using Retrieval-Augmented Fine-Tuning (RAFT). Its core function is to automate the generation of OpenFOAM 11 boundary condition (BC) files through a unique template-patch methodology. The model operates within a RAG pipeline, where it receives a reference tutorial BC file and a natural-language simulation description, then outputs a JSON patch specification to adapt the template's boundary conditions to the new case.
Key Capabilities
- Automated BC Generation: Generates JSON patch specs for OpenFOAM 11 boundary condition files (
0/U,0/p,0/k,0/epsilon,0/omega,0/nuTilda). - Retrieval-Augmented Fine-Tuning (RAFT): Utilizes a RAFT dataset comprising Oracle (50%), Cross-geometry (30%), and Distractor (20%) records, enabling robust adaptation and domain knowledge application.
- Chain-of-Thought Reasoning: Provides step-by-step reasoning within
<thinking>tags before the JSON output, explaining BC assignments. - Supported Turbulence Models: Handles
laminar,kOmegaSST,kEpsilon, andSpalartAllmarasturbulence models. - High Context Length: Supports a maximum sequence length of 32768 tokens, allowing for comprehensive input contexts.
Good for
- OpenFOAM Case Setup Automation: Ideal for developers and engineers looking to streamline the setup of OpenFOAM simulations by automating the creation of boundary condition files.
- Integrating into RAG Pipelines: Designed to be a component of a larger RAG pipeline for automatic OpenFOAM case generation, from mesh to simulation.
- Reducing Manual Configuration: Significantly reduces the manual effort and potential for errors in configuring complex OpenFOAM boundary conditions.
- Educational and Research Purposes: Useful for understanding how LLMs can be fine-tuned for highly specialized, domain-specific code generation tasks.