Skip to content

Commit d567f0e

Browse files
author
Swetha Mandava
committed
mismatch numbers
1 parent 23d56b7 commit d567f0e

File tree

1 file changed

+9
-9
lines changed

1 file changed

+9
-9
lines changed

TensorFlow/LanguageModeling/BERT/README.md

Lines changed: 9 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -1102,14 +1102,14 @@ BERT BASE FP16
11021102

11031103
| Sequence Length | Batch Size | Throughput-Average(sent/sec) | Throughput speedup (FP32 to mixed precision) | Latency-Average(ms) | Latency-90%(ms) | Latency-95%(ms) | Latency-99%(ms) |
11041104
|-----------------|------------|------------------------------|----------------------------------------------|---------------------|-----------------|-----------------|-----------------|
1105-
| 128 | 1 | 13.22 | 1.552234557 | 18.34 | 19.09 | 19.28 | 21.74 |
1106-
| 128 | 2 | 12.97 | 2.521498285 | 20.92 | 21.86 | 22.61 | 23.33 |
1107-
| 128 | 4 | 13.21 | 3.434760186 | 30.03 | 30.32 | 30.43 | 31.06 |
1108-
| 128 | 8 | 18.82 | 4.352926012 | 47.38 | 48.21 | 48.56 | 49.25 |
1109-
| 384 | 1 | 15.44 | 2.87008547 | 29.78 | 30.3 | 30.46 | 31.69 |
1110-
| 384 | 2 | 19.38 | 3.576623377 | 48.41 | 49.03 | 49.26 | 50.04 |
1111-
| 384 | 4 | 30.84 | 3.94635373 | 84.96 | 86.88 | 87.38 | 88.3 |
1112-
| 384 | 8 | 56.31 | 4.254885302 | 159.76 | 162.37 | 163.23 | 165.79 |
1105+
| 128 | 1 | 95.19 | 1.48 | 10.51 | 12.71 | 13.10 | 13.22 |
1106+
| 128 | 2 | 190.68 | 1.72 | 10.49 | 11.16 | 12.85 | 12.97 |
1107+
| 128 | 4 | 339.48 | 2.70 | 11.78 | 11.82 | 11.90 | 13.21 |
1108+
| 128 | 8 | 456.50 | 3.58 | 17.52 | 17.88 | 18.08 | 18.82 |
1109+
| 384 | 1 | 86.57 | 2.44 | 11.55 | 11.83 | 12.95 | 15.44 |
1110+
| 384 | 2 | 118.30 | 3.18 | 16.91 | 16.91 | 17.08 | 19.38 |
1111+
| 384 | 4 | 140.11 | 3.80 | 28.55 | 28.89 | 29.23 | 30.84 |
1112+
| 384 | 8 | 150.84 | 4.18 | 53.04 | 54.58 | 55.19 | 56.31 |
11131113

11141114
BERT BASE FP32
11151115

@@ -1152,4 +1152,4 @@ March 2019
11521152
### Known issues
11531153

11541154

1155-
- There is a known performance regression with the 19.08 release on Tesla V100 boards with 16 GB memory, smaller batch sizes may be a better choice for this model on these GPUs with the 19.08 release. 32 GB GPUs are not affected.
1155+
- There is a known performance regression with the 19.08 release on Tesla V100 boards with 16 GB memory, smaller batch sizes may be a better choice for this model on these GPUs with the 19.08 release. 32 GB GPUs are not affected.

0 commit comments

Comments
 (0)