maxmustermann/DeepLearningExamples
1
0
Fork 0
Code Issues Pull requests Projects Releases Wiki Activity

[Transformer-XL/TF] Updated perf table

Browse source
This commit is contained in:
Krzysztof Kudrynski 2020-07-09 11:09:15 +02:00
parent 3337f72cff
commit 40c3be6e9b
1 changed files with 5 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

8
TensorFlow/LanguageModeling/Transformer-XL/README.md
View file

@ -40,7 +40,6 @@ to achieve state-of-the-art accuracy and is tested and maintained by NVIDIA.
* [Training accuracy: NVIDIA DGX-2 (16x V100 32G)](#training-accuracy-nvidia-dgx-2-16x-v100-32gb)
* [Base model](#base-model-2)
* [Training loss plot](#training-loss-plot)
* [Base model](#base-model-3)
* [Training stability test](#training-stability-test)
* [Base model](#base-model-4)
* [Training performance results](#training-performance-results)
@ -893,8 +892,11 @@ training iterations.
|**GPUs**|**Batch Size / GPU**|**Throughput - TF32 (tok/s)**|**Throughput - Mixed precision (tok/s)**|**Throughput speedup (TF32 to Mixed precision)**|**Weak Scaling - TF32**|**Weak Scaling - Mixed precision**|
|-------:|-------------------:|----------------------------:|---------------------------------------:|-----------------------------------------------:|----------------------:|---------------------------------:|
| 1 | 16 | 34,244 | 36,455 | 1.065 | 1.000 | 6.555 |
| 8 | 32 | 224,474 | 227,502 | 1.013 | 8.636 | 6.241 |
| 1 | 16 | 25,127 | 26,130 | 1.040 | 1.000 | 1.000 |
| 1 | 32 | 30,958 | 33,117 | 1.070 | 1.000 | 1.000 |
| 1 | 64 | 34,244 | 36,455 | 1.065 | 1.000 | 1.000 |
| 8 | 16 | 157,538 | 155,656 | 0.988 | 6.270 | 5.957 |
| 8 | 32 | 224,474 | 227,502 | 1.013 | 7.251 | 6.870 |
To achieve these same results, follow the steps in the [Quick Start Guide](#quick-start-guide).