|
|
--- |
|
|
library_name: transformers |
|
|
tags: |
|
|
- symbolic-math |
|
|
- sympy |
|
|
- equation-classification |
|
|
- text-classification |
|
|
- math-assistant |
|
|
- math-ml |
|
|
- experimental |
|
|
license: mit |
|
|
--- |
|
|
|
|
|
# 🔢 Symbolic Math Expression Classifier (Experimental) |
|
|
|
|
|
This model is an **early-stage classifier** for **SymPy-style symbolic expressions**. |
|
|
It predicts the type of symbolic mathematical input so that the it |
|
|
can route the query to the appropriate symbolic solver (convexity, derivatives, integrals, equations, systems, etc.). |
|
|
|
|
|
👉 This model lives in the development branch **`ml-classify`** on github |
|
|
and is still under active experimentation. |
|
|
|
|
|
> ⚠️ **This is NOT production-ready. |
|
|
> Accuracy is still low on complex expressions and multi-variable cases. |
|
|
> Bugs and misclassifications are expected.** |
|
|
|
|
|
--- |
|
|
|
|
|
# Model Details |
|
|
|
|
|
## Model Description |
|
|
|
|
|
This model takes **pure SymPy syntax strings** as input and classifies them into categories like: |
|
|
|
|
|
- `expression` |
|
|
- `equation` |
|
|
- `system_of_equations` |
|
|
- `linear_equation` |
|
|
- `nonlinear_equation` |
|
|
- `derivative` (e.g., `diff(x**3, x)`) |
|
|
- `integral` (e.g., `Integral(sin(x), x)`) |
|
|
- `convexity_problem` |
|
|
|
|
|
It is trained on **synthetic symbolic data** generated using SymPy templates. |
|
|
|
|
|
### Metadata |
|
|
|
|
|
- **Developed for:** Math Verification of LLM response |
|
|
- **Finetuned from:** RoBERTa (depending on your notebook) |
|
|
- **Language:** English + SymPy tokens |
|
|
- **Format:** Plain text |
|
|
- **License:** MIT |
|
|
|
|
|
--- |
|
|
|
|
|
# Model Sources |
|
|
|
|
|
- **Repository:** https://github.com/math_neuro-ao |
|
|
- **Training Notebook:** [symbolic_classifier.ipynb](https://colab.research.google.com/drive/1ONuLHDp8Y93U_RYXAwJANP8-FD4kpxmX#scrollTo=hxWHsGjJM1Wx) |
|
|
- **Branch:** `ml-classify` |
|
|
|
|
|
--- |
|
|
|
|
|
# Intended Uses |
|
|
|
|
|
## Direct Use |
|
|
Load the classifier like any HuggingFace model: |
|
|
|
|
|
```python |
|
|
from transformers import AutoTokenizer, AutoModelForSequenceClassification |
|
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained("<your-username>/symbolic-math-classifier") |
|
|
model = AutoModelForSequenceClassification.from_pretrained("<your-username>/symbolic-math-classifier") |
