GLUE evaluation automation script

benchmarks/glue.py

@@ -0,0 +1,191 @@
+# 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.
+
+"""GLUE benchmark script to test model performance.
+
+To run the script, use this command:
+```
+python3 glue.py --model BertClassifier \
+                --preset bert_base_en \
+                --epochs 5 \
+                --batch_size 16 \
+                --learning_rate 0.001 \
+                --mixed_precision_policy mixed_float16
+```
+
+Disclaimer: This script only supports GLUE/mrpc (for now).
+"""
+
+import inspect
+import time
+
+import tensorflow as tf
+import tensorflow_datasets as tfds
+from absl import app
+from absl import flags
+from absl import logging
+from tensorflow import keras
+
+import keras_nlp
+
+seed = 42
+tf.random.set_seed(seed)
+
+
+flags.DEFINE_string(
+    "task",
+    "mrpc",
+    "The name of the GLUE task to finetune on.",
+)
+flags.DEFINE_string(
+    "model", None, "The name of the classifier such as BertClassifier."
+)
+flags.DEFINE_string(
+    "preset",
+    None,
+    "The model preset, e.g., 'bert_base_en_uncased' for `BertClassifier`",
+)
+flags.DEFINE_float(
+    "learning_rate", 0.005, "The learning_rate for the optimizer."
+)
+flags.DEFINE_string(
+    "mixed_precision_policy",
+    "mixed_float16",
+    "The global precision policy to use, e.g., 'mixed_float16' or 'float32'.",
+)
+flags.DEFINE_integer("epochs", 2, "The number of epochs.")
+flags.DEFINE_integer("batch_size", 8, "Batch Size.")
+
+
+FLAGS = flags.FLAGS
+
+
+def load_data():
+    """Load data.
+
+    Load GLUE/MRPC dataset, and convert the dictionary format to
+    (features, label), where `features` is a tuple of all input sentences.
+    """
+    feature_names = ("sentence1", "sentence2")
+
+    def split_features(x):
+        # GLUE comes with dictonary data, we convert it to a uniform format
+        # (features, label), where features is a tuple consisting of all
+        # features.
+        features = tuple([x[name] for name in feature_names])
+        label = x["label"]
+        return (features, label)
+
+    train_ds, test_ds, validation_ds = tfds.load(
+        "glue/mrpc",
+        split=["train", "test", "validation"],
+    )
+
+    train_ds = train_ds.map(split_features, num_parallel_calls=tf.data.AUTOTUNE)
+    test_ds = test_ds.map(split_features, num_parallel_calls=tf.data.AUTOTUNE)
+    validation_ds = validation_ds.map(
+        split_features, num_parallel_calls=tf.data.AUTOTUNE
+    )
+    return train_ds, test_ds, validation_ds
+
+
+def load_model(model, preset, num_classes):
+    for name, symbol in keras_nlp.models.__dict__.items():
+        if inspect.isclass(symbol) and issubclass(symbol, keras.Model):
+            if model and name != model:
+                continue
+            if not hasattr(symbol, "from_preset"):
+                continue
+            for preset in symbol.presets:
+                if preset and preset != preset:
+                    continue
+                model = symbol.from_preset(preset, num_classes=num_classes)
+                logging.info(f"\nUsing model {name} with preset {preset}\n")
+                return model
+
+    raise ValueError(f"Model {model} or preset {preset} not found.")
+
+
+def main(_):
+    keras.mixed_precision.set_global_policy(FLAGS.mixed_precision_policy)
+
+    # Check task is supported.
+    # TODO(chenmoneygithub): Add support for other glue tasks.
+    if FLAGS.task != "mrpc":
+        raise ValueError(
+            f"For now only mrpc is supported, but received {FLAGS.task}."
+        )
+
+    logging.info(
+        "Benchmarking configs...\n"
+        "=========================\n"
+        f"MODEL: {FLAGS.model}\n"
+        f"PRESET: {FLAGS.preset}\n"
+        f"TASK: glue/{FLAGS.task}\n"
+        f"BATCH_SIZE: {FLAGS.batch_size}\n"
+        f"EPOCHS: {FLAGS.epochs}\n"
+        "=========================\n"
+    )
+
+    # Load datasets.
+    train_ds, test_ds, validation_ds = load_data()
+    train_ds = train_ds.batch(FLAGS.batch_size).prefetch(tf.data.AUTOTUNE)
+    test_ds = test_ds.batch(FLAGS.batch_size).prefetch(tf.data.AUTOTUNE)
+    validation_ds = validation_ds.batch(FLAGS.batch_size).prefetch(
+        tf.data.AUTOTUNE
+    )
+
+    # Load the model.
+    model = load_model(model=FLAGS.model, preset=FLAGS.preset, num_classes=2)
+    # Set loss and metrics.
+    loss = keras.losses.SparseCategoricalCrossentropy(from_logits=True)
+    metrics = [keras.metrics.SparseCategoricalAccuracy()]
+    # Configure optimizer.
+    lr = tf.keras.optimizers.schedules.PolynomialDecay(
+        FLAGS.learning_rate,
+        decay_steps=train_ds.cardinality() * FLAGS.epochs,
+        end_learning_rate=0.0,
+    )
+    optimizer = tf.keras.optimizers.experimental.AdamW(lr, weight_decay=0.01)
+    optimizer.exclude_from_weight_decay(
+        var_names=["LayerNorm", "layer_norm", "bias"]
+    )
+    model.compile(optimizer=optimizer, loss=loss, metrics=metrics)
+
+    # Start training.
+    logging.info("Starting Training...")
+
+    st = time.time()
+    history = model.fit(
+        train_ds,
+        validation_data=validation_ds,
+        epochs=FLAGS.epochs,
+    )
+
+    wall_time = time.time() - st
+    validation_accuracy = history.history["val_sparse_categorical_accuracy"][-1]
+    examples_per_second = (
+        FLAGS.epochs * FLAGS.batch_size * (len(train_ds) + len(validation_ds))
+    ) / wall_time
+
+    logging.info("Training Finished!")
+    logging.info(f"Wall Time: {wall_time:.4f} seconds.")
+    logging.info(f"Validation Accuracy: {validation_accuracy:.4f}")
+    logging.info(f"examples_per_second: {examples_per_second:.4f}")
+
+
+if __name__ == "__main__":
+    flags.mark_flag_as_required("model")
+    flags.mark_flag_as_required("preset")
+    app.run(main)