cs-552-2026-ma-que/general_knowledge_model

TEXT GENERATIONConcurrency Cost:1Model Size:2BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:May 17, 2026Architecture:Transformer Cold

The cs-552-2026-ma-que/general_knowledge_model is a 1.7 billion parameter language model developed by Ma Que, fine-tuned from Qwen/Qwen3-1.7B-Base. It is specifically designed for general knowledge multiple-choice answering, providing reasoning-style responses with a final boxed answer. This model excels at academic benchmark tasks requiring structured, reasoned answers to general knowledge questions.

Loading preview...

Overview

The cs-552-2026-ma-que/general_knowledge_model is a specialized language model developed by Ma Que for the EPFL CS-552 course. It is based on the Qwen3-1.7B-Base model and has undergone LoRA SFT (Supervised Fine-Tuning) to excel in general knowledge multiple-choice answering. The model is designed to produce reasoning traces before committing to a final, single-letter boxed answer, adhering to a specific output contract for automated parsing.

Key Capabilities

  • General Knowledge Multiple-Choice Answering: Optimized to read questions with lettered options and select the correct one.
  • Reasoning-Style Responses: Trained to generate internal thought processes (<think>...</think>) before providing the final answer.
  • Structured Output: Emits the final answer in the format \boxed{X} for compatibility with parsers like OpenCompass/vLLM.
  • LoRA Fine-Tuning: Utilizes LoRA with rank 256 and alpha 256, targeting key attention and feed-forward projection layers for efficient adaptation.

Intended Use & Limitations

This model is primarily a research artifact for the CS-552 general knowledge benchmark. It is intended for academic evaluation and producing structured multiple-choice answers. Users should be aware that it is not an authoritative factual system and may contain outdated, incomplete, or incorrect information. Outputs require verification before any real-world or safety-critical applications.