New Update in Upcoming Noor Ul Ilm 2.0 22-03-2026
Noor ul Ilm 2.0
Reasoning Verifier
The world's first self-correcting Islamic AI — now thinking, verifying, and refining before every answer
Introducing the Reasoning Verifier Pipeline
Noor ul Ilm 2.0 introduces a structured reasoning pipeline that teaches the model to think carefully, verify its own answer, and only respond once it is confident. This is a new approach — not found in other Islamic AI models or general AI models currently available.
<verify> catches a mistake → loops back and checks again → repeats until the answer is fully correct → only then moves to <end>
This means the model does not give up until it gets the answer right. Especially for Islamic knowledge — where accuracy matters — this approach ensures greater reliability before the final response is shown to the user.
Setting Expectations
As with all fine-tuned models, Noor ul Ilm 2.0 is a work in progress. The model may not always follow the <think> <verify> <end> pattern perfectly — it may sometimes hallucinate or skip steps. This is normal and expected at this stage of fine-tuning. With each training session, the model becomes more familiar with the pipeline and improves. Insha Allah, future versions will follow it more consistently.
We believe in being transparent. This is an open, community-driven project — not a commercial product. Your feedback and patience are truly appreciated. Jazakumullahu Khairan.
Model Name
Built on top of the Noor ul Ilm 1.0 base (Qwen2.5-7B fine-tuned on Quran, Hadith, Tafsir, and Fatawa), this version adds the reasoning verifier training on top via a multi-session relay fine-tuning system on Google Colab. Release coming soon insha Allah.
Access Noor ul Ilm
All models are completely free and open source — for the ummah, by the ummah.
Comments
Post a Comment