Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Light-weight benchmarking script #664

Merged
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
24 changes: 24 additions & 0 deletions keras_nlp/benchmarks/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -25,3 +25,27 @@ the following results were obtained:
To change the configuration, say, for example, number of layers in the transformer
model used for inference, the user can modify the config dictionaries given at
the top of the script.

mattdangerw marked this conversation as resolved.
Show resolved Hide resolved
## Sentiment Analysis

For benchmarking classification models, the following command can be run
from the root of the repository:

```sh
python3 keras_nlp/benchmarks/sentiment_analysis.py \
--model="BertClassifier" \
--preset="bert_small_en_uncased" \
--learning_rate=5e-5 \
--num_epochs=5 \
--batch_size=32
--mixed_precision_policy="mixed_float16"
```

flag `--model` specifies the model name, and `--preset` specifies the preset under testing. `--preset` could be None,
while `--model` is required. Other flags are common training flags.

This script outputs:

- validation accuracy for each epoch.
- testing accuracy after training is done.
- total elapsed time (in seconds).
mattdangerw marked this conversation as resolved.
Show resolved Hide resolved
144 changes: 144 additions & 0 deletions keras_nlp/benchmarks/sentiment_analysis.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,144 @@
# Copyright 2023 The KerasNLP Authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import inspect
import time

import tensorflow as tf
import tensorflow_datasets as tfds
from absl import app
from absl import flags
from tensorflow import keras

import keras_nlp

FLAGS = flags.FLAGS
flags.DEFINE_string(
"model",
None,
"The name of the classifier such as BertClassifier.",
)
flags.DEFINE_string(
"preset",
None,
"The name of a preset, e.g. bert_base_multi.",
)

flags.DEFINE_string(
"mixed_precision_policy",
"mixed_float16",
"The global precision policy to use. E.g. 'mixed_float16' or 'float32'.",
)

flags.DEFINE_float("learning_rate", 5e-5, "The learning rate.")
flags.DEFINE_integer("num_epochs", 1, "The number of epochs.")
flags.DEFINE_integer("batch_size", 16, "The batch size.")

tfds.disable_progress_bar()

BUFFER_SIZE = 10000


def create_imdb_dataset():
dataset, info = tfds.load(
"imdb_reviews", as_supervised=True, with_info=True
)
train_dataset, test_dataset = dataset["train"], dataset["test"]

train_dataset = (
train_dataset.shuffle(BUFFER_SIZE)
.batch(FLAGS.batch_size)
.prefetch(tf.data.AUTOTUNE)
)

# We split the test data evenly into validation and test sets.
test_dataset_size = info.splits["test"].num_examples // 2

val_dataset = (
test_dataset.take(test_dataset_size)
.batch(FLAGS.batch_size)
.prefetch(tf.data.AUTOTUNE)
)
test_dataset = (
test_dataset.skip(test_dataset_size)
.batch(FLAGS.batch_size)
.prefetch(tf.data.AUTOTUNE)
)

return train_dataset, val_dataset, test_dataset


def create_model():
for name, symbol in keras_nlp.models.__dict__.items():
if inspect.isclass(symbol) and issubclass(symbol, keras.Model):
if FLAGS.model and name != FLAGS.model:
continue
if not hasattr(symbol, "from_preset"):
continue
for preset in symbol.presets:
if FLAGS.preset and preset != FLAGS.preset:
continue
model = symbol.from_preset(preset)
print(f"Using model {name} with preset {preset}")
return model

raise ValueError(f"Model {FLAGS.model} or preset {FLAGS.preset} not found.")


def train_model(
model: keras.Model,
train_dataset: tf.data.Dataset,
validation_dataset: tf.data.Dataset,
):
model.compile(
loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
optimizer=keras.optimizers.Adam(5e-5),
metrics=keras.metrics.SparseCategoricalAccuracy(),
jit_compile=True,
)

model.fit(
train_dataset,
epochs=FLAGS.num_epochs,
validation_data=validation_dataset,
verbose=2,
)

return model


def evaluate_model(model: keras.Model, test_dataset: tf.data.Dataset):
loss, accuracy = model.evaluate(test_dataset)
print(f"Test loss: {loss}")
print(f"Test accuracy: {accuracy}")


def main(_):
keras.mixed_precision.set_global_policy(FLAGS.mixed_precision_policy)

# Start time
start_time = time.time()

train_dataset, validation_dataset, test_dataset = create_imdb_dataset()
model = create_model()
model = train_model(model, train_dataset, validation_dataset)
evaluate_model(model, test_dataset)

# End time
end_time = time.time()
print(f"Total wall time: {end_time - start_time}")


if __name__ == "__main__":
flags.mark_flag_as_required("model")
app.run(main)