llleb/mistral-7b-arc-qlora-exp7-3

TEXT GENERATIONConcurrent Unit Cost:1Model Size:7BQuant:FP8Context Size:4kTool Calling:SupportedPublished:Jul 2, 2026Architecture:Transformer Featherless Exclusive Cold

The llleb/mistral-7b-arc-qlora-exp7-3 is a 7 billion parameter Mistral-7B-v0.1 model fine-tuned using QLoRA for the ARC-Challenge science multiple-choice question answering task. It was trained with a 4-bit NF4 QLoRA method and response-only loss on a dataset including ARC-Challenge, ARC-Easy, and OpenBookQA. This model is specifically optimized for scientific reasoning and factual recall in a multiple-choice format, demonstrating its strength in academic QA benchmarks.

Loading preview...

Model Overview

llleb/mistral-7b-arc-qlora-exp7-3 is a 7 billion parameter model based on mistralai/Mistral-7B-v0.1. It has been specifically fine-tuned using the QLoRA method to excel at the ARC-Challenge science multiple-choice question answering task. This model leverages 4-bit NF4 QLoRA with response-only loss, making it efficient for deployment while maintaining strong performance in its specialized domain.

Key Capabilities

  • Specialized QA: Optimized for science multiple-choice question answering, particularly on the ARC-Challenge dataset.
  • Efficient Fine-tuning: Utilizes QLoRA for efficient adaptation of the base Mistral-7B model.
  • Targeted Training Data: Trained on a combination of ARC-Challenge, ARC-Easy (subset), and OpenBookQA datasets to enhance its scientific reasoning abilities.

Evaluation

The model's performance was evaluated using lm-evaluation-harness on the arc_challenge benchmark, employing a 25-shot prompting strategy. This evaluation confirms its proficiency in handling complex scientific questions in a multiple-choice format.

Good For

  • Academic Question Answering: Ideal for tasks requiring factual recall and reasoning in scientific domains.
  • Educational Tools: Can be integrated into systems for generating or answering science-related quizzes and tests.
  • Research in QLoRA: Provides a practical example of QLoRA's effectiveness in task-specific fine-tuning for QA.