---
language: en
tags:
- azbert
- pretraining
- fill-mask
widget:
- text: "Determine the [MASK] of $f$ $($ $x$ [MASK] $equal$ $x$ $plus$ $root$ ${$ $4$ $minus$ $x$ $supscript$ $2$ $}$ without [MASK]"
  example_title: "mathy"
- text: "Solve $y$ $prime$ [MASK] $minus$ $4$ $y$ $prime$ $plus$ $4$ $y$ [MASK] $x$ $e$ $supscript$ $x$"
  example_title: "mathy"
- text: "$f$ $($ $x$ [MASK] $y$ $)$"
  example_title: "mathy"
- text: "$x$ [MASK] $x$ $equal$ $2$ $x$"
  example_title: "mathy"
- text: "With Euler's [MASK], it [MASK] to $int$ $subscript$ $0$ [MASK] [MASK] $frac$ ${$ $1$ $plus$ $x$ $supscript$ $2$ $}$ ${$ $1$ $plus$ $x$ $}$ $d$ $x$"
  example_title: "mathy"
- text: "Proof by [MASK] that $n$ [MASK] $gt$ $3$ $n$ [MASK] $n$ [MASK] $6$"
  example_title: "mathy"
license: mit
---

## About
Here we share a pretrained BERT model that is aware of math tokens. The math tokens are treated specially and tokenized using [pya0](https://github.com/approach0/pya0), which adds very limited new tokens for latex markup (total vocabulary is just 31,061).

This model is trained on 4 x 2 Tesla V100 with a total batch size of 64, using Math StackExchange data with 2.7 million sentence pairs trained for 7 epochs.

### Usage
Download and try it out
```sh
pip install pya0==0.3.2
wget https://vault.cs.uwaterloo.ca/s/gqstFZmWHCLGXe3/download -O ckpt.tar.gz
mkdir -p ckpt
tar xzf ckpt.tar.gz -C ckpt --strip-components=1
python test.py --test_file test.txt
```

### Test file format
Modify the test examples in `test.txt` to play with it.

The test file is tab-separated, the first column is additional positions you want to mask for the right-side sentence (useful for masking tokens in math markups). A zero means no additional mask positions.

### Example output
![](https://i.imgur.com/xpl87KO.png)

### Upload to huggingface
This repo is hosted on [Github](https://github.com/approach0/azbert), and only mirrored at [huggingface](https://huggingface.co/castorini/azbert-base).

To upload to huggingface, use the `upload2hgf.sh` script.
Before runnig this script, be sure to check:
* check points for model and tokenizer are created under `./ckpt` folder
* model contains all the files needed: `config.json` and `pytorch_model.bin`
* tokenizer contains all the files needed: `added_tokens.json`, `special_tokens_map.json`, `tokenizer_config.json`, `vocab.txt` and `tokenizer.json`
* no `tokenizer_file` field in `tokenizer_config.json` (sometimes it is located locally at `~/.cache`)
* `git-lfs` is installed
* having git-remote named `hgf` reference to `https://huggingface.co/castorini/azbert-base`