SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Doc styling (#8067)

Browse Source 
* Important files

* Styling them all

* Revert "Styling them all"

This reverts commit 7d029395fdae8513b8281cbc2a6c239f8093503e.

* Syling them for realsies

* Fix syntax error

* Fix benchmark_utils

* More fixes

* Fix modeling auto and script

* Remove new line

* Fixes

* More fixes

* Fix more files

* Style

* Add FSMT

* More fixes

* More fixes

* More fixes

* More fixes

* Fixes

* More fixes

* More fixes

* Last fixes

* Make sphinx happy
This commit is contained in:
Sylvain Gugger
2020-10-26 18:26:02 -04:00
committed by GitHub
parent 04a17f8550
commit 08f534d2da
271 changed files with 9726 additions and 8991 deletions
Download Patch File Download Diff File Expand all files Collapse all files

101
docs/source/benchmarks.rst
View File

@@ -3,21 +3,27 @@ Benchmarks
Let's take a look at how 🤗 Transformer models can be benchmarked, best practices, and already available benchmarks.
A notebook explaining in more detail how to benchmark 🤗 Transformer models can be found `here <https://github.com/huggingface/transformers/blob/master/notebooks/05-benchmark.ipynb>`__.
A notebook explaining in more detail how to benchmark 🤗 Transformer models can be found `here
<https://github.com/huggingface/transformers/blob/master/notebooks/05-benchmark.ipynb>`__.
How to benchmark 🤗 Transformer models
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The classes :class:`~transformers.PyTorchBenchmark` and :class:`~transformers.TensorFlowBenchmark` allow to flexibly benchmark 🤗 Transformer models.
The benchmark classes allow us to measure the `peak memory usage` and `required time` for both
`inference` and `training`.
The classes :class:`~transformers.PyTorchBenchmark` and :class:`~transformers.TensorFlowBenchmark` allow to flexibly
benchmark 🤗 Transformer models. The benchmark classes allow us to measure the `peak memory usage` and `required time`
for both `inference` and `training`.
.. note::
Hereby, `inference` is defined by a single forward pass, and `training` is defined by a single forward pass and backward pass.
Hereby, `inference` is defined by a single forward pass, and `training` is defined by a single forward pass and
backward pass.
The benchmark classes :class:`~transformers.PyTorchBenchmark` and :class:`~transformers.TensorFlowBenchmark` expect an object of type :class:`~transformers.PyTorchBenchmarkArguments` and :class:`~transformers.TensorFlowBenchmarkArguments`, respectively, for instantiation. :class:`~transformers.PyTorchBenchmarkArguments` and :class:`~transformers.TensorFlowBenchmarkArguments` are data classes and contain all relevant configurations for their corresponding benchmark class.
In the following example, it is shown how a BERT model of type `bert-base-cased` can be benchmarked.
The benchmark classes :class:`~transformers.PyTorchBenchmark` and :class:`~transformers.TensorFlowBenchmark` expect an
object of type :class:`~transformers.PyTorchBenchmarkArguments` and
:class:`~transformers.TensorFlowBenchmarkArguments`, respectively, for instantiation.
:class:`~transformers.PyTorchBenchmarkArguments` and :class:`~transformers.TensorFlowBenchmarkArguments` are data
classes and contain all relevant configurations for their corresponding benchmark class. In the following example, it
is shown how a BERT model of type `bert-base-cased` can be benchmarked.
.. code-block::
@@ -34,11 +40,15 @@ In the following example, it is shown how a BERT model of type `bert-base-cased`
>>> benchmark = TensorFlowBenchmark(args)
Here, three arguments are given to the benchmark argument data classes, namely ``models``, ``batch_sizes``, and ``sequence_lengths``. The argument ``models`` is required and expects a :obj:`list` of model identifiers from the `model hub <https://huggingface.co/models>`__
The :obj:`list` arguments ``batch_sizes`` and ``sequence_lengths`` define the size of the ``input_ids`` on which the model is benchmarked.
There are many more parameters that can be configured via the benchmark argument data classes. For more detail on these one can either directly consult the files
``src/transformers/benchmark/benchmark_args_utils.py``, ``src/transformers/benchmark/benchmark_args.py`` (for PyTorch) and ``src/transformers/benchmark/benchmark_args_tf.py`` (for Tensorflow).
Alternatively, running the following shell commands from root will print out a descriptive list of all configurable parameters for PyTorch and Tensorflow respectively.
Here, three arguments are given to the benchmark argument data classes, namely ``models``, ``batch_sizes``, and
``sequence_lengths``. The argument ``models`` is required and expects a :obj:`list` of model identifiers from the
`model hub <https://huggingface.co/models>`__ The :obj:`list` arguments ``batch_sizes`` and ``sequence_lengths`` define
the size of the ``input_ids`` on which the model is benchmarked. There are many more parameters that can be configured
via the benchmark argument data classes. For more detail on these one can either directly consult the files
``src/transformers/benchmark/benchmark_args_utils.py``, ``src/transformers/benchmark/benchmark_args.py`` (for PyTorch)
and ``src/transformers/benchmark/benchmark_args_tf.py`` (for Tensorflow). Alternatively, running the following shell
commands from root will print out a descriptive list of all configurable parameters for PyTorch and Tensorflow
respectively.
.. code-block:: bash
@@ -65,7 +75,7 @@ An instantiated benchmark object can then simply be run by calling ``benchmark.r
bert-base-uncased 8 128 0.018
bert-base-uncased 8 512 0.088
--------------------------------------------------------------------------------
==================== INFERENCE - MEMORY - RESULT ====================
--------------------------------------------------------------------------------
Model Name Batch Size Seq Length Memory in MB
@@ -75,7 +85,7 @@ An instantiated benchmark object can then simply be run by calling ``benchmark.r
bert-base-uncased 8 128 1307
bert-base-uncased 8 512 1539
--------------------------------------------------------------------------------
==================== ENVIRONMENT INFORMATION ====================
- transformers_version: 2.11.0
- framework: PyTorch
@@ -98,7 +108,7 @@ An instantiated benchmark object can then simply be run by calling ``benchmark.r
- gpu_power_watts: 280.0
- gpu_performance_state: 2
- use_tpu: False
>>> ## TENSORFLOW CODE
>>> results = benchmark.run()
>>> print(results)
@@ -111,7 +121,7 @@ An instantiated benchmark object can then simply be run by calling ``benchmark.r
bert-base-uncased 8 128 0.022
bert-base-uncased 8 512 0.105
--------------------------------------------------------------------------------
==================== INFERENCE - MEMORY - RESULT ====================
--------------------------------------------------------------------------------
Model Name Batch Size Seq Length Memory in MB
@@ -121,7 +131,7 @@ An instantiated benchmark object can then simply be run by calling ``benchmark.r
bert-base-uncased 8 128 1330
bert-base-uncased 8 512 1770
--------------------------------------------------------------------------------
==================== ENVIRONMENT INFORMATION ====================
- transformers_version: 2.11.0
- framework: Tensorflow
@@ -145,14 +155,17 @@ An instantiated benchmark object can then simply be run by calling ``benchmark.r
- gpu_performance_state: 2
- use_tpu: False
By default, the `time` and the `required memory` for `inference` are benchmarked.
In the example output above the first two sections show the result corresponding to `inference time` and `inference memory`.
In addition, all relevant information about the computing environment, `e.g.` the GPU type, the system, the library versions, etc... are printed out in the third section under `ENVIRONMENT INFORMATION`.
This information can optionally be saved in a `.csv` file when adding the argument :obj:`save_to_csv=True` to :class:`~transformers.PyTorchBenchmarkArguments` and :class:`~transformers.TensorFlowBenchmarkArguments` respectively.
In this case, every section is saved in a separate `.csv` file. The path to each `.csv` file can optionally be defined via the argument data classes.
By default, the `time` and the `required memory` for `inference` are benchmarked. In the example output above the first
two sections show the result corresponding to `inference time` and `inference memory`. In addition, all relevant
information about the computing environment, `e.g.` the GPU type, the system, the library versions, etc... are printed
out in the third section under `ENVIRONMENT INFORMATION`. This information can optionally be saved in a `.csv` file
when adding the argument :obj:`save_to_csv=True` to :class:`~transformers.PyTorchBenchmarkArguments` and
:class:`~transformers.TensorFlowBenchmarkArguments` respectively. In this case, every section is saved in a separate
`.csv` file. The path to each `.csv` file can optionally be defined via the argument data classes.
Instead of benchmarking pre-trained models via their model identifier, `e.g.` `bert-base-uncased`, the user can alternatively benchmark an arbitrary configuration of any available model class.
In this case, a :obj:`list` of configurations must be inserted with the benchmark args as follows.
Instead of benchmarking pre-trained models via their model identifier, `e.g.` `bert-base-uncased`, the user can
alternatively benchmark an arbitrary configuration of any available model class. In this case, a :obj:`list` of
configurations must be inserted with the benchmark args as follows.
.. code-block::
@@ -183,7 +196,7 @@ In this case, a :obj:`list` of configurations must be inserted with the benchmar
bert-6-lay 8 128 0.009
bert-6-lay 8 512 0.044
--------------------------------------------------------------------------------
==================== INFERENCE - MEMORY - RESULT ====================
--------------------------------------------------------------------------------
Model Name Batch Size Seq Length Memory in MB
@@ -201,7 +214,7 @@ In this case, a :obj:`list` of configurations must be inserted with the benchmar
bert-6-lay 8 128 1127
bert-6-lay 8 512 1359
--------------------------------------------------------------------------------
==================== ENVIRONMENT INFORMATION ====================
- transformers_version: 2.11.0
- framework: PyTorch
@@ -252,7 +265,7 @@ In this case, a :obj:`list` of configurations must be inserted with the benchmar
bert-6-lay 8 128 0.0011
bert-6-lay 8 512 0.074
--------------------------------------------------------------------------------
==================== INFERENCE - MEMORY - RESULT ====================
--------------------------------------------------------------------------------
Model Name Batch Size Seq Length Memory in MB
@@ -270,7 +283,7 @@ In this case, a :obj:`list` of configurations must be inserted with the benchmar
bert-6-lay 8 128 1330
bert-6-lay 8 512 1540
--------------------------------------------------------------------------------
==================== ENVIRONMENT INFORMATION ====================
- transformers_version: 2.11.0
- framework: Tensorflow
@@ -295,8 +308,9 @@ In this case, a :obj:`list` of configurations must be inserted with the benchmar
- use_tpu: False
Again, `inference time` and `required memory` for `inference` are measured, but this time for customized configurations of the :obj:`BertModel` class. This feature can especially be helpful when
deciding for which configuration the model should be trained.
Again, `inference time` and `required memory` for `inference` are measured, but this time for customized configurations
of the :obj:`BertModel` class. This feature can especially be helpful when deciding for which configuration the model
should be trained.
Benchmark best practices
@@ -304,19 +318,28 @@ Benchmark best practices
This section lists a couple of best practices one should be aware of when benchmarking a model.
- Currently, only single device benchmarking is supported. When benchmarking on GPU, it is recommended that the user
specifies on which device the code should be run by setting the ``CUDA_VISIBLE_DEVICES`` environment variable in the shell, `e.g.` ``export CUDA_VISIBLE_DEVICES=0`` before running the code.
- The option :obj:`no_multi_processing` should only be set to :obj:`True` for testing and debugging. To ensure accurate memory measurement it is recommended to run each memory benchmark in a separate process by making sure :obj:`no_multi_processing` is set to :obj:`True`.
- One should always state the environment information when sharing the results of a model benchmark. Results can vary heavily between different GPU devices, library versions, etc., so that benchmark results on their own are not very useful for the community.
- Currently, only single device benchmarking is supported. When benchmarking on GPU, it is recommended that the user
specifies on which device the code should be run by setting the ``CUDA_VISIBLE_DEVICES`` environment variable in the
shell, `e.g.` ``export CUDA_VISIBLE_DEVICES=0`` before running the code.
- The option :obj:`no_multi_processing` should only be set to :obj:`True` for testing and debugging. To ensure accurate
memory measurement it is recommended to run each memory benchmark in a separate process by making sure
:obj:`no_multi_processing` is set to :obj:`True`.
- One should always state the environment information when sharing the results of a model benchmark. Results can vary
heavily between different GPU devices, library versions, etc., so that benchmark results on their own are not very
useful for the community.
Sharing your benchmark
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Previously all available core models (10 at the time) have been benchmarked for `inference time`, across many different settings: using PyTorch, with
and without TorchScript, using TensorFlow, with and without XLA. All of those tests were done across CPUs (except for
TensorFlow XLA) and GPUs.
Previously all available core models (10 at the time) have been benchmarked for `inference time`, across many different
settings: using PyTorch, with and without TorchScript, using TensorFlow, with and without XLA. All of those tests were
done across CPUs (except for TensorFlow XLA) and GPUs.
The approach is detailed in the `following blogpost <https://medium.com/huggingface/benchmarking-transformers-pytorch-and-tensorflow-e2917fb891c2>`__ and the results are available `here <https://docs.google.com/spreadsheets/d/1sryqufw2D0XlUH4sq3e9Wnxu5EAQkaohzrJbd5HdQ_w/edit?usp=sharing>`__.
The approach is detailed in the `following blogpost
<https://medium.com/huggingface/benchmarking-transformers-pytorch-and-tensorflow-e2917fb891c2>`__ and the results are
available `here
<https://docs.google.com/spreadsheets/d/1sryqufw2D0XlUH4sq3e9Wnxu5EAQkaohzrJbd5HdQ_w/edit?usp=sharing>`__.
With the new `benchmark` tools, it is easier than ever to share your benchmark results with the community `here <https://github.com/huggingface/transformers/blob/master/examples/benchmarking/README.md>`__.
With the new `benchmark` tools, it is easier than ever to share your benchmark results with the community `here
<https://github.com/huggingface/transformers/blob/master/examples/benchmarking/README.md>`__.