About This Project

Contextual-Sentence-Creator is a fine-tuned language model based on IBM Granite 4.2 3B, developed to produce synthetic training data for Automatic Speech Recognition (ASR) systems. Given a specific term, a context description, and a requested quantity, the model generates natural English sentences ranging from 6 to 18 words, ensuring the target term is used exactly once in the provided context. This tool is specifically optimized for creating high-quality, domain-specific vocabulary lists, including technical terms, names, and specialized cultural terminology.

Built using LoRA fine-tuning on a custom dataset of over 13,000 examples, the model is designed for programmatic integration and returns output as a structured JSON array. It is particularly useful for developers and researchers who need to generate localized speech-to-text datasets at scale. While the model is highly effective at following formatting constraints, it is intended for generating plausible, readable sentences rather than factually verified statements.

Project Details

Tags

transformerssafetensorsgranitetext-generationsentence-generationasrvocabularyloraconversationalendataset:danielrosehill/term-context-sentencesbase_model:ibm-granite/granite-4.2-3bbase_model:adapter:ibm-granite/granite-4.2-3blicense:apache-2.0endpoints_compatibleregion:us

Explore More Projects