Update README.md
Browse files
README.md
CHANGED
@@ -10,7 +10,7 @@ tags:
|
|
10 |
|
11 |
lyraBELLE is currently the **fastest BELLE model** available. To the best of our knowledge, it is the **first accelerated version of BELLE**.
|
12 |
|
13 |
-
The inference speed of lyraBELLE has achieved **
|
14 |
|
15 |
Among its main features are:
|
16 |
|
@@ -29,21 +29,18 @@ Note that:
|
|
29 |
|
30 |
### test environment
|
31 |
|
32 |
-
**it takes a few minutes to load the model and much longer time when you use original version. Just be patient..**
|
33 |
-
|
34 |
- device: Nvidia A100 40G
|
35 |
- warmup: 10 rounds
|
36 |
- percision:fp16
|
37 |
-
- batch size
|
38 |
-
- batch size for the original: xx (almost maximum under A100 40G)
|
39 |
- language: chinese, keep same in a batch.
|
|
|
40 |
|
41 |
|
42 |
-
|version|
|
43 |
|:-:|:-:|:-:|
|
44 |
-
|original|
|
45 |
-
|lyraBELLE|
|
46 |
-
|lyraBELLE|96|3507 tokens/sec|
|
47 |
|
48 |
|
49 |
|
@@ -90,7 +87,7 @@ print(output_texts)
|
|
90 |
``` bibtex
|
91 |
@Misc{lyraBELLE2023,
|
92 |
author = {Kangjian Wu, Zhengtao Wang, Bin Wu},
|
93 |
-
title = {lyraBELLE: Accelerating BELLE by
|
94 |
howpublished = {\url{https://huggingface.co/TMElyralab/lyraBELLE},
|
95 |
year = {2023}
|
96 |
}
|
|
|
10 |
|
11 |
lyraBELLE is currently the **fastest BELLE model** available. To the best of our knowledge, it is the **first accelerated version of BELLE**.
|
12 |
|
13 |
+
The inference speed of lyraBELLE has achieved **3x+** acceleration upon the original version.
|
14 |
|
15 |
Among its main features are:
|
16 |
|
|
|
29 |
|
30 |
### test environment
|
31 |
|
|
|
|
|
32 |
- device: Nvidia A100 40G
|
33 |
- warmup: 10 rounds
|
34 |
- percision:fp16
|
35 |
+
- batch size:64
|
|
|
36 |
- language: chinese, keep same in a batch.
|
37 |
+
- do_sample: True, model generate slightly different answser to same question.
|
38 |
|
39 |
|
40 |
+
|version|speed|
|
41 |
|:-:|:-:|:-:|
|
42 |
+
|original|826.34 tokens/sec|
|
43 |
+
|lyraBELLE|2701.71 tokens/sec|
|
|
|
44 |
|
45 |
|
46 |
|
|
|
87 |
``` bibtex
|
88 |
@Misc{lyraBELLE2023,
|
89 |
author = {Kangjian Wu, Zhengtao Wang, Bin Wu},
|
90 |
+
title = {lyraBELLE: Accelerating BELLE by 3x+},
|
91 |
howpublished = {\url{https://huggingface.co/TMElyralab/lyraBELLE},
|
92 |
year = {2023}
|
93 |
}
|