akilx/qwen-english-mcq

TEXT GENERATIONConcurrency Cost:1Model Size:2BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Jun 3, 2026License:mitArchitecture:Transformer0.0K Open Weights Cold

akilx/qwen-english-mcq is a 2 billion parameter causal language model developed by Akil Al-Sharafi and team, based on Qwen3-1.7B. It is specifically fine-tuned using Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) for high-quality, instruction-less Multiple Choice Question (MCQ) generation from English documents. This model excels at automatically generating educational assessments and exam questions by completing XML-formatted prompts.

Loading preview...

Overview

akilx/qwen-english-mcq is a 2 billion parameter causal language model developed by Akil Al-Sharafi, Jihad Nassari, Bashar Alwan, Adeeb Hassan, and Akram AL-subari. It is built upon the Qwen/Qwen3-1.7B base model and has been meticulously fine-tuned using a combination of Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) via Unsloth. This specialized training makes it highly effective for generating Multiple Choice Questions (MCQs) from English text.

Key Capabilities

  • Instruction-less MCQ Generation: Unlike general-purpose models, it operates as a pure causal text completer, generating MCQs without explicit natural language instructions.
  • XML-based Prompting: It expects input text within <context>...</context> tags, followed by an opening <question> tag, and then auto-completes the question, answer, and distractors, stopping at </distractors>.
  • Optimized for Educational Assessment: Its primary domain is the automatic generation of high-quality MCQs for educational purposes and exam creation.
  • Compact and Efficient: Leveraging a 1.7B parameter base model, it offers efficient performance for its specialized task.

How it Works

The model is designed for causal auto-completion. Users provide a text segment within <context> tags, followed by <question>, and the model generates the complete MCQ structure. This methodology is detailed in an accompanying research paper (focused on an Arabic counterpart) which explores bridging performance gaps in MCQ generation using compact models and DPO/RLAIF techniques.

Finetunes
1 models