psdba/deepseek-mcq-trainer-mohammedsayeeduddin-lora

DeepSeek MCQ Trainer (LoRA) — by Mohammed Sayeeduddin

Model Details

• Model name: psdba/deepseek-mcq-trainer-mohammedsayeeduddin-lora
• Developed by: Mohammed Sayeeduddin
• Model type: LoRA adapter (PEFT) fine-tuned for structured MCQ generation
• Base model: deepseek-ai/deepseek-coder-1.3b-instruct
• Primary use: Instructor-grade MCQ generation in strict JSON format for IT training
• Language(s): English
• License: This repository contains LoRA adapter weights only. Usage is subject to the license of the base model (deepseek-ai/deepseek-coder-1.3b-instruct). Please review and comply with that license before commercial or redistribution use.

Model Description

This is a specialist training model designed to generate multiple-choice questions (MCQs) in a strict, machine-readable JSON schema.

The adapter was fine-tuned to produce:

• 4 options (A/B/C/D)
• exactly 1 correct answer
• short explanation
• JSON-only responses (no markdown, no extra commentary)

Intended Use

Direct Use

• Corporate training MCQs (FastAPI, Docker/Linux, Python Core, LLM/RAG)
• Classroom quizzes and practice tests
• Building MCQ datasets for LMS/Excel ingestion

Downstream Use

• Integration into training platforms (MCQ generators, exam portals)
• Dataset generation pipelines (JSON → CSV → LMS)

Out-of-Scope Use

• Medical / legal / financial advice
• Open-domain chat or creative writing
• High-stakes decisions without human validation

Bias, Risks, and Limitations

• MCQs may still contain imperfections or ambiguous distractors.
• Always validate questions before real exams or certifications.
• Model can hallucinate if prompts are unclear or request unsupported topics.

How to Get Started

Install

pip install -U transformers peft accelerate