Exploring different approaches to customize language models for domain-specific text-to-code generation

ArXi:2603.16526v1 Announce Type: new Large language models (LLMs) have nstrated strong capabilities in generating executable code from natural language descriptions. However, general-purpose models often struggle in specialized programming contexts where domain-specific libraries, APIs, or conventions must be used. Customizing smaller open-source models offers a cost-effective alternative to relying on large