From 9dfb9bf2dd93fb6f28f6fddf7e1e714a59828fac Mon Sep 17 00:00:00 2001 From: Christopher Akiki Date: Thu, 15 Jun 2023 14:36:38 +0200 Subject: [PATCH 1/2] [MINOR:TYPO] Update README.md --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index 5393af0..7badc13 100644 --- a/README.md +++ b/README.md @@ -51,7 +51,7 @@ We have precompiled some popular models and listed with the source code referenc | roberta.bin | Byte-BPE tokenization model for Roberta model | byte BPE | [src](https://github.com/microsoft/BlingFire/tree/master/ldbsrc/roberta) | | syllab.bin | Multi lingual model to identify allowed hyphenation points inside a word. | W2H | [src](https://github.com/microsoft/BlingFire/tree/master/ldbsrc/syllab) | -Oh yes, it is also the fastest! We did a comparison of Bling Fire with tokenizers from Hugging Face, [Bling Fire runs 4-5 times faster than Hugging Face Tokenizers](https://github.com/Microsoft/BlingFire/wiki/Comparing-performance-of-Bling-Fire-and-Hugging-Face-Tokenizers), see also [Bing Blog Post](https://blogs.bing.com/Developers-Blog/march-2020/Bling-FIRE-Tokenizer-for-BERT). We did comparison of Bling Fire Unigram LM and BPE implementaion to the same one in [SentencePiece](https://github.com/google/sentencepiece) library and our implementation is ~2x faster, see [XLNET benchmark](https://github.com/microsoft/BlingFire/blob/master/ldbsrc/xlnet/README.TXT) and [BPE benchmark](https://github.com/microsoft/BlingFire/blob/master/ldbsrc/bpe_example/README.TXT). Not to mention our default models are 10x faster than the same functionality from [SpaCy](https://github.com/explosion/spaCy), see [benchmark wiki](https://github.com/Microsoft/BlingFire/wiki/Benchmark-Guide) and this [Bing Blog Post](https://blogs.bing.com/Developers-Blog/2019-04/bling-fire-tokenizer-released-to-open-source). +Oh yes, it is also the fastest! We did a comparison of Bling Fire with tokenizers from Hugging Face, [Bling Fire runs 4-5 times faster than Hugging Face Tokenizers](https://github.com/Microsoft/BlingFire/wiki/Comparing-performance-of-Bling-Fire-and-Hugging-Face-Tokenizers), see also [Bing Blog Post](https://blogs.bing.com/Developers-Blog/march-2020/Bling-FIRE-Tokenizer-for-BERT). We did comparison of Bling Fire Unigram LM and BPE implementation to the same one in [SentencePiece](https://github.com/google/sentencepiece) library and our implementation is ~2x faster, see [XLNET benchmark](https://github.com/microsoft/BlingFire/blob/master/ldbsrc/xlnet/README.TXT) and [BPE benchmark](https://github.com/microsoft/BlingFire/blob/master/ldbsrc/bpe_example/README.TXT). Not to mention our default models are 10x faster than the same functionality from [SpaCy](https://github.com/explosion/spaCy), see [benchmark wiki](https://github.com/Microsoft/BlingFire/wiki/Benchmark-Guide) and this [Bing Blog Post](https://blogs.bing.com/Developers-Blog/2019-04/bling-fire-tokenizer-released-to-open-source). So if low latency inference is what you need then you have to try Bling Fire! From b3252f238efd5eabccdca46bf965c5625909dfe1 Mon Sep 17 00:00:00 2001 From: Christopher Akiki Date: Thu, 15 Jun 2023 14:38:05 +0200 Subject: [PATCH 2/2] [MINOR:TYPO] Update README.md --- dist-pypi/README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/dist-pypi/README.md b/dist-pypi/README.md index 778740b..be0abaa 100644 --- a/dist-pypi/README.md +++ b/dist-pypi/README.md @@ -45,7 +45,7 @@ We have precompiled some popular models and listed with the source code referenc | roberta.bin | Byte-BPE tokenization model for Roberta model | byte BPE | [src](https://github.com/microsoft/BlingFire/tree/master/ldbsrc/roberta) | | syllab.bin | Multi lingual model to identify allowed hyphenation points inside a word. | W2H | [src](https://github.com/microsoft/BlingFire/tree/master/ldbsrc/syllab) | -Oh yes, it is also the fastest! We did a comparison of Bling Fire with tokenizers from Hugging Face, [Bling Fire runs 4-5 times faster than Hugging Face Tokenizers](https://github.com/Microsoft/BlingFire/wiki/Comparing-performance-of-Bling-Fire-and-Hugging-Face-Tokenizers), see also [Bing Blog Post](https://blogs.bing.com/Developers-Blog/march-2020/Bling-FIRE-Tokenizer-for-BERT). We did comparison of Bling Fire Unigram LM and BPE implementaion to the same one in [SentencePiece](https://github.com/google/sentencepiece) library and our implementation is ~2x faster, see [XLNET benchmark](https://github.com/microsoft/BlingFire/blob/master/ldbsrc/xlnet/README.TXT) and [BPE benchmark](https://github.com/microsoft/BlingFire/blob/master/ldbsrc/bpe_example/README.TXT). Not to mention our default models are 10x faster than the same functionality from [SpaCy](https://github.com/explosion/spaCy), see [benchmark wiki](https://github.com/Microsoft/BlingFire/wiki/Benchmark-Guide) and this [Bing Blog Post](https://blogs.bing.com/Developers-Blog/2019-04/bling-fire-tokenizer-released-to-open-source). +Oh yes, it is also the fastest! We did a comparison of Bling Fire with tokenizers from Hugging Face, [Bling Fire runs 4-5 times faster than Hugging Face Tokenizers](https://github.com/Microsoft/BlingFire/wiki/Comparing-performance-of-Bling-Fire-and-Hugging-Face-Tokenizers), see also [Bing Blog Post](https://blogs.bing.com/Developers-Blog/march-2020/Bling-FIRE-Tokenizer-for-BERT). We did comparison of Bling Fire Unigram LM and BPE implementation to the same one in [SentencePiece](https://github.com/google/sentencepiece) library and our implementation is ~2x faster, see [XLNET benchmark](https://github.com/microsoft/BlingFire/blob/master/ldbsrc/xlnet/README.TXT) and [BPE benchmark](https://github.com/microsoft/BlingFire/blob/master/ldbsrc/bpe_example/README.TXT). Not to mention our default models are 10x faster than the same functionality from [SpaCy](https://github.com/explosion/spaCy), see [benchmark wiki](https://github.com/Microsoft/BlingFire/wiki/Benchmark-Guide) and this [Bing Blog Post](https://blogs.bing.com/Developers-Blog/2019-04/bling-fire-tokenizer-released-to-open-source). So if low latency inference is what you need then you have to try Bling Fire!