grade6-math-assistant

base_model: Qwen/Qwen2.5-Math-1.5B peft tags:

math unsloth qwen qwen2.5 lora synthetic-data grade-6

license: apache-2.0 language: en

Grade 6 Math Assistant (LoRA)

This is a fine-tuned LoRA adapter designed to solve Grade 6 Math Olympiad problems. It generates step-by-step reasoning (Chain of Thought) before arriving at a final standalone answer.

Model Details

Base Model: Qwen/Qwen2.5-1.5B Training Method: LoRA (Low-Rank Adaptation) via Unsloth. Dataset: Synthetically generated Grade 6 math olympiad problems (focused on fractions, ratios, early algebra, and combinatorics).

Intended Use

This model is designed to take a single math problem and output a step-by-step logical breakdown followed by the final answer.

Prompt Format: For the model to work correctly, you MUST format your prompt exactly like this:

Below is a math problem. Solve it step-by-step.

Problem:

[Insert your math problem here]

Reasoning:

Known Limitations Single Objective Only: This model was trained on problems requiring a single final answer. It currently struggles with multi-part questions (e.g., "Part A" and "Part B"). Please split multi-part questions into separate prompts. Base Model Requirement: Because this is a LoRA adapter, you cannot run it by itself. It must be merged with the base Qwen2.5-1.5B model at runtime.

Training Infrastructure

Trained using Unsloth for 2x faster training.

Dataset generated using OpenAI gpt-4o-2024-08-06 Structured Outputs.