Hi, we are a team at Microsoft called Bling (Beyond Language Understanding), we help Bing be smarter. Here we wanted to share with all of you our FInite State machine and REgular expression manipulation library (FIRE). We use Fire for many linguistic operations inside Bing such as Tokenization, Multi-word expression matching, Unknown word-guessing, Stemming / Lemmatization just to mention a few.
Bling Fire Tokenizer is a tokenizer designed for fast-speed and quality tokenization of Natural Language text.
The default model follows the tokenization logic of NLTK, except hyphenated words are split and a few errors are fixed. We also support arbitrary models with normalization and sub-token extraction like in BERT tokenizer. The tokenizer high level API designed in a way that it requires minimal or no configuration, or initialization, or additional files and is friendly for use from languages like Python, Perl, C#, Java, etc.
The default model is built in and does not need a file to load.
We provide 4 compiled models for BERT base/large, BERT base/large cased, BERT Chinese and BERT Multilinugual Cased.
File Name | Models it should be used for |
---|---|
bert_base_tok.bin | BERT Base/Large |
bert_base_cased_tok.bin | BERT Base/Large Cased |
bert_chinese.bin | BERT Chinese |
bert_multi_cased.bin | BERT Multi Lingual Cased |
Oh yes, it is also fast, as uses minimal deterministic finite state machines underneath. So if real-time inference is what you are doing and latency is an issue you have to try Bling Fire!
NLTK:
The South Florida/Miami area has previously hosted the event 10 times . Names such as French , ( De ) Roche , Devereux , D'Arcy , Treacy and Lacy are particularly common in the southeast of Ireland . Marconi 's European experiments in July 1899—Marconi may have transmitted the letter S ( dot/dot/dot ) in a naval demonstration In the confirmation window , click OK. Review the FMT Real - time Report ES . Go to C : \Users\Public\Documents\hyper - v\Virtual hard disks\ and delete MSIT_Win10.VHDX . … and an agency / vendor company are regulated by the country ' s civil code ; labor relationships between a …
FIRE:
The South Florida / Miami area has previously hosted the event 10 times . Names such as French , ( De ) Roche , Devereux , D' Arcy , Treacy and Lacy are particularly common in the southeast of Ireland . Marconi 's European experiments in July 1899 — Marconi may have transmitted the letter S ( dot / dot / dot ) in a naval demonstration In the confirmation window , click OK . Review the FMT Real - time Report ES . Go to C : \ Users \ Public \ Documents \ hyper - v \ Virtual hard disks \ and delete MSIT_Win10 . VHDX . … and an agency / vendor company are regulated by the country 's civil code ; labor relationships between a …
Comparing Bling Fire with other popular NLP libraries, Bling Fire shows 10X faster speed in tokenization task
System | Avg Run Time (Second Per 10,000 Passages) |
---|---|
Bling Fire | 0.823 |
SpaCy | 8.653 |
NLTK | 17.821 |
See more at benchmark wiki
If you simply want to use it in Python, you can install the latest release using pip:
pip install -U blingfire
from blingfire import *
text = 'This is the Bling-Fire tokenizer'
output = text_to_words(text)
print(output)
This is the Bling - Fire tokenizer
On one thread, it works 14x faster than orignal BERT tokenizer written in Python. Given this code is written in C++ it can be called from multiple threads without blocking on global interpreter lock thus achiving higher speed-ups for batch mode.
import os
import blingfire
s = "Эpple pie. How do I renew my virtual smart card?: /Microsoft IT/ 'virtual' smart card certificates for DirectAccess are valid for one year. In order to get to microsoft.com we need to type [email protected]."
# one time load the model (we are using the one that comes with the package)
h = blingfire.load_model(os.path.join(os.path.dirname(blingfire.__file__), "bert_base_tok.bin"))
print("Model Handle: %s" % h)
# use the model from one or more threads
print(s)
ids = blingfire.text_to_ids(h, s, 128, 100) # sequence length: 128, oov id: 100
print(ids) # returns a numpy array of length 128 (padded or trimmed)
print(s+s)
ids = blingfire.text_to_ids(h, s+s, 128, 100)
print(ids)
# free the model at the end
blingfire.free_model(h)
print("Model Freed")
Model Handle: 2854016629088
Эpple pie. How do I renew my virtual smart card?: /Microsoft IT/ 'virtual' smart card certificates for DirectAccess are valid for one year. In order to get to microsoft.com we need to type [email protected].
[ 1208 9397 2571 11345 1012 2129 2079 1045 20687 2026 7484 6047
4003 1029 1024 1013 7513 2009 1013 1005 7484 1005 6047 4003
17987 2005 3622 6305 9623 2015 2024 9398 2005 2028 2095 1012
1999 2344 2000 2131 2000 7513 1012 4012 2057 2342 2000 2828
14255 1030 1015 1012 1016 1012 1015 1012 1016 1012 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0]
Эpple pie. How do I renew my virtual smart card?: /Microsoft IT/ 'virtual' smart card certificates for DirectAccess are valid for one year. In order to get to microsoft.com we need to type [email protected].Эpple pie. How do I renew my virtual smart card?: /Microsoft IT/ 'virtual' smart card certificates for DirectAccess are valid for one year. In order to get to microsoft.com we need to type [email protected].
[ 1208 9397 2571 11345 1012 2129 2079 1045 20687 2026 7484 6047
4003 1029 1024 1013 7513 2009 1013 1005 7484 1005 6047 4003
17987 2005 3622 6305 9623 2015 2024 9398 2005 2028 2095 1012
1999 2344 2000 2131 2000 7513 1012 4012 2057 2342 2000 2828
14255 1030 1015 1012 1016 1012 1015 1012 1016 1012 1208 9397
2571 11345 1012 2129 2079 1045 20687 2026 7484 6047 4003 1029
1024 1013 7513 2009 1013 1005 7484 1005 6047 4003 17987 2005
3622 6305 9623 2015 2024 9398 2005 2028 2095 1012 1999 2344
2000 2131 2000 7513 1012 4012 2057 2342 2000 2828 14255 1030
1015 1012 1016 1012 1015 1012 1016 1012 0 0 0 0
0 0 0 0 0 0 0 0]
Model Freed
This notebook demonstrates how Bling Fire tokenizer helps in Stack Overflow posts classification problem.
If you want to create your own tokenization or any other finite-state model, you need to compile the C++ tools first. Then use these tools to compile linugusitc resources from human readble format into binary finite-state machines.
- This Document is an introduction into compilation of the C++ code and linguistic resources.
- Adding BERT-like tokenization model is describing how to add new tokenization model similar to BERT.
This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.microsoft.com.
When you submit a pull request, a CLA-bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., label, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.
This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact [email protected] with any additional questions or comments.
To contribute directly to code base, you should create a personal fork and create feature branches there when you need them. This keeps the main repository clean and your personal workflow out of sight.
Before we can accept a pull request from you, you'll need to sign a Contributor License Agreement (CLA). It is an automated process and you only need to do it once.
However, you don't have to do this up-front. You can simply clone, fork, and submit your pull-request as usual. When your pull-request is created, it is classified by a CLA bot. If the change is trivial (i.e. you just fixed a typo) then the PR is labelled with cla-not-required
. Otherwise, it's classified as cla-required
. In that case, the system will also tell you how you can sign the CLA. Once you have signed a CLA, the current and all future pull-requests will be labelled as cla-signed
.
To enable us to quickly review and accept your pull requests, always create one pull request per issue and link the issue in the pull request if possible. Never merge multiple requests in one unless they have the same root cause. Besides, keep code changes as small as possible and avoid pure formatting changes to code that has not been modified otherwise.
- Ask a question on Stack Overflow.
- File a bug in GitHub Issues.
Security issues and bugs should be reported privately, via email, to the Microsoft Security Response Center (MSRC) at [email protected]. You should receive a response within 24 hours. If for some reason you do not, please follow up via email to ensure we received your original message. Further information, including the MSRC PGP key, can be found in the Security TechCenter.
Copyright (c) Microsoft Corporation. All rights reserved.
Licensed under the MIT License.