kenhktsui / math-fasttext-classifier

maths-fasttext-classifier

Dataset

This is part of my fasttext classifier collection for curating pretraining dataset. This classifier classifies a text into Maths or Others.
The model is trained over 1.6M records, which is a 50:50 mix of maths and non maths in website and achieved a test F1 score of 0.99 (too good to be true?). It is an intended upsampling of maths data. The classifier can be used for LLM pretraining data curation, to enhance capability in mathematics. It is ultra fast ⚡ with a throughtput of ~2000 doc/s with CPU.

Don't underestimate the "old" fasttext classiifer! It is indeed a good and scalable practice. For example, QWEN2.5-MATH leverages fasttext to curate pretraining data, althought its classifier is not open sourced.

🛠️Usage

from typing import List
import re
from huggingface_hub import hf_hub_download
import fasttext


model_hf = fasttext.load_model(hf_hub_download("kenhktsui/maths-fasttext-classifier", "model.bin"))


def replace_newlines(text: str) -> str:
  return re.sub("\n+", " ", text)


def predict(text_list: List[str]) -> List[dict]:
  text_list = [replace_newlines(text) for text in text_list]
  pred = model.predict(text_list)
  return [{"label": l[0].lstrip("__label__"), "score": s[0]}
           for l, s in zip(*pred)]


predict([
  """This is a lightning fast model, which can classify at throughtput of 2000 doc/s with CPU""",
  """Differential geometry is a mathematical discipline that studies the geometry of smooth shapes and smooth spaces, otherwise known as smooth manifolds. It uses the techniques of single variable calculus, vector calculus, linear algebra and multilinear algebra.""",
  """Given $p$: $|4x-3|\leqslant 1$ and $q$: $x^{2}-(2a+1)x+a^{2}+a\leqslant 0$, find the range of values for $a$ if $p$ is a necessary but not sufficient condition for $q$."""
])
# [{'label': 'Others', 'score': 1.00000834},
# {'label': 'Maths', 'score': 0.99995351},
# {'label': 'Maths', 'score': 0.99801832}]

📊Evaluation

full version

              precision    recall  f1-score   support

       Maths       0.99      0.98      0.99    200000
      Others       0.98      0.99      0.99    200000

    accuracy                           0.99    400000
   macro avg       0.99      0.99      0.99    400000
weighted avg       0.99      0.99      0.99    400000

⚠️Known Limitation

The classifier does not handle short text well, which might not be surprising.