MidTool-fasttext-pdf-quality-classifier

The fastText quality classifier used to filter PDF documents when building MidTool-Mix. It scores a document on whether it is useful tool-use / developer-facing technical material (manuals, product handbooks, platform documentation) rather than generic or noisy PDF-extraction output.

It is the PDF counterpart of MidTool-fasttext-web-quality-classifier, trained the same way but applied with a stricter threshold, since PDF extraction is noisier than web text.

Usage

import fasttext
from huggingface_hub import hf_hub_download

path = hf_hub_download("MidTool/MidTool-fasttext-pdf-quality-classifier", "model.bin")
model = fasttext.load_model(path)
labels, probs = model.predict(text.replace("\n", " "))

Input should be a single line of whitespace-normalized text.

Details

See our paper for the full data, training, and evaluation details.

@article{jiang2026midtool,
  title  = {MidTool: Mid-training Data Synthesis for Agentic Tool Use},
  author = {Jiang, Fengqing and Wang, Yite and Liu, Boyi and Wang, Zhaoyang and
            Xu, Canwen and Yao, Zhewei and Poovendran, Radha and He, Yuxiong},
  year   = {2026}
}