Inspired by a similar model, this model addresses the same challenge: providing an efficient way to generate questions from raw data. It is highly versatile and requires no specific data formatting; it is even robust enough to handle noisy or low-quality OCR text.

Designed for:

Golden dataset generation
RAG benchmarking
Generating HyDE indices for QnA systems
Evaluation corpus bootstrapping
Retriever quality testing
Graph RAG generation

The model is trained to generate unstructured output consisting of a single atomic question. Due to its small scale, it may struggle to produce correctly formatted structured data (e.g., JSON).

Suggested prompt template

Given the text below, extract ONE question grounded strictly in a single atomic fact.

<text>
<your_text_here>
</text>

Return ONLY the question:

Uploaded finetuned model

Developed by: Catlilface
License: apache-2.0
Finetuned from model : Catlilface/Qwen3.5-0.8B-interrogator

This qwen3_5 model was trained 2x faster with Unsloth and Huggingface's TRL library.