Update README.md

README.md

@@ -10,7 +10,7 @@ tags:
 
 lyraBELLE is currently the **fastest BELLE model** available. To the best of our knowledge, it is the **first accelerated version of BELLE**.
 
-The inference speed of lyraBELLE has achieved **100x** acceleration upon the original version.
+The inference speed of lyraBELLE has achieved **3x+** acceleration upon the original version.
 
 Among its main features are:
 
@@ -29,21 +29,18 @@ Note that:
 
 ### test environment
 
-**it takes a few minutes to load the model and much longer time when you use original version. Just be patient..**
-
 - device: Nvidia A100 40G
 - warmup: 10 rounds
 - percision：fp16
-- batch size for our version: 96 (almost maximum under A100 40G)
-- batch size for the original: xx (almost maximum under A100 40G)
+- batch size：64
 - language： chinese, keep same in a batch.
+- do_sample: True, model generate slightly different answser to same question.
 
 
-|version|batch size|speed|
+|version|speed|
 |:-:|:-:|:-:|
-|original|50|xx|
-|lyraBELLE|50|xx|
-|lyraBELLE|96|3507 tokens/sec|
+|original|826.34 tokens/sec|
+|lyraBELLE|2701.71 tokens/sec|
 
 
 
@@ -90,7 +87,7 @@ print(output_texts)
 ``` bibtex
 @Misc{lyraBELLE2023,
   author =       {Kangjian Wu, Zhengtao Wang, Bin Wu},
-  title =        {lyraBELLE: Accelerating BELLE by 100x+},
+  title =        {lyraBELLE: Accelerating BELLE by 3x+},
   howpublished = {\url{https://huggingface.co/TMElyralab/lyraBELLE},
   year =         {2023}
 }