ptvnck/qwen2.5-1.5b-exam-tutor

TEXT GENERATIONConcurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 9, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

ptvnck/qwen2.5-1.5b-exam-tutor is a 1.5 billion parameter Qwen2.5-Instruct model fine-tuned by ptvnck with a 32768 token context length. It is specifically designed to act as a patient human tutor, guiding users through exam preparation by asking probing questions rather than directly providing answers. This model excels at conversational tutoring for subjects like math word problems and conceptual explanations, focusing on developing critical thinking.

Loading preview...

Qwen2.5-1.5B Exam Tutor: A Conversational Tutoring Assistant

qwen2.5-1.5b-exam-tutor is a LoRA fine-tune of the Qwen2.5-1.5B-Instruct model, developed by ptvnck. Unlike general-purpose LLMs, this model is specialized to function as a patient human tutor, guiding students to think through problems rather than simply providing solutions. It achieves this by asking guiding questions and probing for misconceptions, mimicking effective teaching methodologies.

Key Capabilities

  • Conversational Tutoring: Engages in dialogue to help users understand and solve problems.
  • Guided Problem Solving: Focuses on leading students to discover solutions independently.
  • Exam Preparation: Ideal for math word problems, conceptual explanations, and step-by-step reasoning practice.
  • Efficient Fine-tuning: Utilizes LoRA (rank 16) with Unsloth and TRL for efficient training on a NVIDIA T4 GPU.
  • Response-Only Loss Masking: Training loss is computed exclusively on tutor turns, optimizing the model's ability to tutor.

Training and Performance

The model was trained on 650 student-tutor dialogues from ptvnck/TutoringDialogs and eth-nlped/mathdial, formatted with the tokenizer's ChatML template. It achieved a best validation loss of 1.506 (perplexity \u2248 4.51) at epoch 3, indicating effective learning of the tutoring pattern. The model is intended as the generation backend for a larger RAG + FastAPI tutoring assistant project.

Good For

  • Integrating into educational applications requiring a conversational tutoring component.
  • Assisting students with exam preparation in a guided, interactive manner.
  • Use cases where the goal is to foster understanding rather than just provide answers.

Finetunes
1 models