A PEFT (LoRA) fine-tuned Mixtral-8x7B-Instruct-v0.1 model specialized for mathematical reasoning, arithmetic, algebra, geometry, statistics and logical problem solving using the Adaption Labs AutoScientist workflow.

Overview

Adaptive Math is a domain-adapted language model developed for mathematical reasoning and instruction following.

The model was fine-tuned using the Adaption Labs AutoScientist workflow on a mathematics reasoning dataset containing thousands of mathematical instruction-completion pairs.

Unlike a general language model, this model focuses on structured mathematical reasoning, symbolic manipulation and logical step-by-step solutions.

Quick Facts

Item	Value
Domain	Mathematics
Task	Mathematical Reasoning
Base Model	Mixtral-8x7B-Instruct-v0.1
Fine-tuning Method	PEFT (LoRA)
Training Framework	Adaption Labs AutoScientist
Dataset	adaption-single-integer-samples
Language	English
License	Apache-2.0

Training Dataset

Dataset Repository

https://huggingface.co/datasets/Charley890/adaption-single-integer-samples

The training dataset contains mathematical reasoning examples covering:

Arithmetic
Algebra
Number Theory
Geometry
Statistics
Logical Reasoning
Symbolic Mathematics
Mathematical Word Problems

Key Features

Mathematical Question Answering
Step-by-Step Reasoning
Arithmetic Computation
Algebraic Manipulation
Geometry Reasoning
Number Theory
Statistics
Symbolic Reasoning
STEM Education
AI Tutoring

Base Model

Property	Value
Base Model	Mixtral-8x7B-Instruct-v0.1
Fine-tuning	PEFT (LoRA)
Framework	Adaption Labs AutoScientist
Language	English

Evaluation Results

Evaluation performed using Adaption Labs AutoScientist

Metric	Base Model	Adapted Model	Improvement
Grade	E	A	Excellent
Overall Win Rate	19	81	+62
Math Win Rate	38	63	+25
Percentile	0.0	57.7	+57.7
Relative Improvement	—	—	82%

Performance Summary

✓ Training Completed Successfully

✓ Stable Optimization

✓ Mathematical Reasoning Improved

✓ Grade Improved from E → A

✓ Overall Win Rate Improved from 19 → 81

✓ Math Win Rate Improved from 38 → 63

✓ Percentile Improved from 0.0 → 57.7

✓ Relative Quality Improvement: 82%

Adaptive Math Demo

Mathematical Knowledge

This model is optimized to solve a broad range of mathematical problems through logical reasoning instead of memorization.

It performs best when explaining every intermediate step before arriving at the final answer.

Supported Topics

Category	Examples
Arithmetic	Addition, Subtraction, Multiplication, Division
Algebra	Linear Equations, Quadratic Equations, Polynomial Factorization
Statistics	Mean, Median, Mode, Variance, Standard Deviation,etc.

Fundamental Mathematical Formulae

Formula	Equation
Quadratic Formula	$x=\frac{-b\pm\sqrt{b^2-4ac}}{2a}$
Pythagorean Theorem	$a^2+b^2=c^2$
Slope Formula	$m=\frac{y_2-y_1}{x_2-x_1}$,etc.

Example Usage

Example 1

Input

text

Solve:
2x+7=19

Output

text

Subtract 7 from both sides.
2x=12
Divide both sides by 2.
x=6

Example 2

Input

text

Find the derivative of
f(x)=x²+5x+2

Output

text

Using the power rule,
f'(x)=2x+5

Example 3

Input

Find the area of a circle with radius 7.

Output

text

A=\pi r^2
A=49\pi
≈153.94


y
10 |                          ●
 9 |                      ●
 8 |                  ●
 7 |              ●
 6 |          ●
 5 |      ●
 4 |   ●
 3 | ●
   +--------------------------------→ x
     1   2   3   4   5   6   7   8

text

Frequency

|                    █
|                    █
|               █    █
|               █    █
|          █    █    █
|     █    █    █    █
|     █    █    █    █
| █   █    █    █    █
| █   █    █    █    █
| █   █    █    █    █
   +----------------------------
      A    B    C    D    E

Inference with Transformers

python

from transformers import AutoTokenizer
from transformers import AutoModelForCausalLM
model_name = "Charlie890/Adaptive-Math-Reasoner-Mixtral-8x7B"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)
prompt = "Solve: 3x + 5 = 20"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(
    **inputs,
    max_new_tokens=256
)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Applications

This model is designed for a wide variety of mathematical and scientific applications.

AI Tutors
STEM Education
Mathematics Chatbots
Scientific Computing
Homework Assistance
Research
Educational Software and more.

Limitations

Although highly capable, the model has several limitations.

May occasionally make arithmetic mistakes on extremely long calculations.
Complex symbolic manipulations should always be verified.
Does not replace professional Computer Algebra Systems (CAS) such as Mathematica or Maple.
Performance depends heavily on prompt quality.
Mathematical proofs may require human verification.
May hallucinate unsupported mathematical identities if prompted incorrectly.
Numerical approximations can accumulate rounding errors.
High-stakes scientific applications should always be independently validated.
Performance may decrease on very large expressions or lengthy derivations.

Future Improvements

Future releases may include:

Better symbolic reasoning
Olympiad-level mathematics
Interactive tutoring
Diagram understanding
Mathematical OCR support
Scientific equation solving and more.

Credit

Adaptive data by Adaption.

Acknowledgements

Special thanks to:

Adaption Labs
Mixtral Team
Hugging Face and Kaggle
Transformers Community
Open-source AI Community
Mathematical research contributors

Citation

If you use this model in your research, please cite:

bibtex

@misc{adaptive_math_reasoner,
  title        = {Adaptive Math Reasoner},
  author       = {Edidiong Charlie},
  year         = {2026},
  publisher    = {Hugging Face},
  model        = {Mixtral-8x7B-Instruct},
  framework    = {Adaption Labs + AutoScientist},
  license      = {Apache-2.0},
  url          = {https://huggingface.co/Charlie890/Adaptive-Math-Reasoner-Mixtral-8x7B}
}

License

This project is released under the Apache License 2.0.

You are free to:

✅ Use commercially
✅ Modify
✅ Distribute
✅ Private use
✅ Research
✅ Education

Subject to the terms and conditions of the Apache License 2.0.

Performance Goals

The model is optimized to:

Produce step-by-step mathematical reasoning.
Solve algebraic equations accurately.
Handle advanced calculus problems.
Solve geometry and trigonometry questions.
Perform statistical computations.
Explain mathematical concepts clearly.
Generate clean LaTeX mathematical expressions.
Assist students, educators, engineers, and researchers.

Example Prompt Ideas

text

Solve:
3x² + 7x - 10 = 0

text

Differentiate:

f(x)=sin(x)e^x

text

Integrate:

∫x²cos(x)dx

text

Find the determinant of

|2 4|
|1 5|

Prove the Binomial Theorem.

Explain Bayes' Theorem with a practical example.

text

Find the eigenvalues of the matrix:

[[4,2],
 [1,3]]

Mission

The goal of Adaptive Math Reasoner is to provide accurate, explainable, and accessible mathematical reasoning powered by modern Large Language Models.

The project aims to make advanced mathematics easier to learn, explore, and apply across education, engineering, science, finance, and research.

Version

Version: 1.0.0

Base Model: Mixtral-8x7B-Instruct

Framework: Transformers

License: Apache-2.0

Primary Domain: Mathematical Reasoning

Author: Edidiong Charlie

Thank You

Thank you for using Adaptive Math Reasoner.

We hope this model helps students, educators, researchers, developers, engineers, and the open-source community solve mathematical problems more effectively.

Happy building with AI and Mathematics! 🚀

Charley890/AdaptiveMath

Adaptive Math — Mixtral-8x7B

Overview

Quick Facts

Training Dataset

Key Features

Base Model

Evaluation Results

Performance Summary

Adaptive Math Demo

Mathematical Knowledge

Supported Topics

Fundamental Mathematical Formulae

Example Usage

Example 1

Example 2

Example 3

Inference with Transformers

Applications

Limitations

Future Improvements

Credit

Acknowledgements

Citation

License

Performance Goals

Example Prompt Ideas

Mission

Version

Thank You

Charley890/
AdaptiveMath