Update benchmark results

chatnoir-eu · Sep 21, 2021 · 991d8ad · 991d8ad
1 parent b587d64
commit 991d8ad
Showing 1 changed file with 4 additions and 4 deletions.
diff --git a/docs/man/parse/lang.rst b/docs/man/parse/lang.rst
@@ -57,14 +57,14 @@ If you know your text is from one of several candidate languages, you can restri
 
 Benchmarks
 ^^^^^^^^^^
-On inputs the size of an average webpage, Resiliparse's fast language detector is about 2--5x as fast as `FastText <https://fasttext.cc/blog/2017/10/02/blog-post.html>`_ (depends on FastText's convergence speed) and even 40x as fast as `langid <https://github.com/saffsd/langid.py>`_:
+On inputs the size of an average webpage, Resiliparse's fast language detector is about 3--5x as fast as `FastText <https://fasttext.cc/blog/2017/10/02/blog-post.html>`_ (depends on FastText's convergence speed) and even 40x as fast as `langid <https://github.com/saffsd/langid.py>`_:
 
 ::
 
   Benchmarking language detectors (10,000 rounds):
-  Resiliparse: 2.4s
-  FastText:    8.4s
-  Langid:      101.3s
+  Resiliparse: 1.7s
+  FastText:    6.8s
+  Langid:      101.7s
 
 Resiliparse's performance advantage comes mostly from the fact that the language detector does not need to tokenize the text or build a vocabulary map at all, which makes it very low-latency, independent of the vocabulary size, and guarantees a fixed memory ceiling and linear runtime complexity.