sadiqoon/sadiq-cite-8b

TEXT GENERATIONPricing:Input $0.468 / Output $1.82Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 2, 2026License:mitArchitecture:Transformer Open Weights Featherless Exclusive Cold

Sadiq-Cite 8B by Sadiqoon Technologies is an 8 billion parameter retrieval-grounded citation model built on Qwen3-8B, specifically designed for Islamic jurisprudence in the Ja'fari (Imami) tradition. It excels at verbatim quotation and citation of rulings from provided passages, rather than generating new verdicts or reasoning. This model is optimized for integration into RAG pipelines, ensuring high accuracy in quoting and citing relevant texts while robustly refusing to improvise when no relevant information is found.

Loading preview...

Sadiq-Cite 8B: Retrieval-Grounded Citation for Islamic Jurisprudence

Sadiq-Cite 8B, developed by Sadiqoon Technologies and based on Qwen3-8B, is a specialized model for Islamic jurisprudence in the Ja'fari (Imami) tradition. Unlike general-purpose LLMs, its core function is to accurately extract and cite relevant rulings from provided passages, quoting them verbatim and attributing them with book and page numbers. It is designed as an answer stage within a retrieval-augmented generation (RAG) pipeline, not a standalone chatbot.

Key Capabilities & Differentiators

  • Verbatim Quotation: Identifies and quotes relevant jurisprudential texts precisely from given passages.
  • Accurate Citation: Provides book and page numbers for each quoted ruling.
  • Refusal to Improvise: Explicitly states when no relevant information is found in the provided passages, preventing hallucination.
  • Machine-Verifiable Output: Designed for a pipeline where every quote can be programmatically verified against the source passages using sadiq_fiqh.verify().
  • Robust Refusal Discrimination: Achieved 100% correct refusals on 100 probes with irrelevant context during evaluation, demonstrating strong grounding.
  • Prompt Format Robustness: Maintains high quoting accuracy across various context injection styles (e.g., AnythingLLM-style, XML tags).

Evaluation Highlights

Evaluated on 282 held-out questions and 100 refusal probes, Sadiq-Cite 8B demonstrated:

  • 99.6% of answers quoted at least one passage.
  • 87.5% of quotes were traceable to supplied passages (84.5% exact, 3.0% near-verbatim).
  • 100% correct refusals on deliberately irrelevant contexts.
  • 0.4% false refusals on genuine questions.

Ideal Use Cases

  • Building specialized RAG systems for Islamic jurisprudence.
  • Automating the extraction and citation of rulings from religious texts.
  • Applications requiring high factual accuracy and verifiability in legal or religious contexts.
  • Integration with tools like LM Studio, AnythingLLM, or Ollama for local deployment within a larger RAG stack, often paired with Fiqh-Embed and Fiqh-Rerank for optimal performance.

Finetunes
2 models