Views
No views yet
pip install -U -q keras-hub
pip install -U -q keras| Preset name | Parameters | Description |
|---|---|---|
| distil_bert_base_en_uncased | 66.36M | 6-layer model where all input is lowercased. |
| distil_bert_base_en | 65.19M | 6-layer model where case is maintained. |
| distil_bert_base_multi | 134.73M | 6-layer multi-linguage model where case is maintained. |
1import keras
2import keras_hub
3import numpy as np1features = ["The quick brown fox jumped.", "I forgot my homework."]
2labels = [0, 3]
3
4# Use a shorter sequence length.
5preprocessor = keras_hub.models.DistilBertPreprocessor.from_preset(
6 "distil_bert_base_en",
7 sequence_length=128,
8)
9# Pretrained classifier.
10classifier = keras_hub.models.DistilBertClassifier.from_preset(
11 "distil_bert_base_en",
12 num_classes=4,
13 preprocessor=preprocessor,
14)
15classifier.fit(x=features, y=labels, batch_size=2)
16
17# Re-compile (e.g., with a new learning rate)
18classifier.compile(
19 loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
20 optimizer=keras.optimizers.Adam(5e-5),
21 jit_compile=True,
22)
23# Access backbone programmatically (e.g., to change `trainable`).
24classifier.backbone.trainable = False
25# Fit again.
26classifier.fit(x=features, y=labels, batch_size=2)1features = {
2 "token_ids": np.ones(shape=(2, 12), dtype="int32"),
3 "padding_mask": np.array([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0]] * 2)
4}
5labels = [0, 3]
6
7# Pretrained classifier without preprocessing.
8classifier = keras_hub.models.DistilBertClassifier.from_preset(
9 "distil_bert_base_en",
10 num_classes=4,
11 preprocessor=None,
12)
13classifier.fit(x=features, y=labels, batch_size=2)1import keras
2import keras_hub
3import numpy as np1features = ["The quick brown fox jumped.", "I forgot my homework."]
2labels = [0, 3]
3
4# Use a shorter sequence length.
5preprocessor = keras_hub.models.DistilBertPreprocessor.from_preset(
6 "hf://keras/distil_bert_base_en",
7 sequence_length=128,
8)
9# Pretrained classifier.
10classifier = keras_hub.models.DistilBertClassifier.from_preset(
11 "hf://keras/distil_bert_base_en",
12 num_classes=4,
13 preprocessor=preprocessor,
14)
15classifier.fit(x=features, y=labels, batch_size=2)
16
17# Re-compile (e.g., with a new learning rate)
18classifier.compile(
19 loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
20 optimizer=keras.optimizers.Adam(5e-5),
21 jit_compile=True,
22)
23# Access backbone programmatically (e.g., to change `trainable`).
24classifier.backbone.trainable = False
25# Fit again.
26classifier.fit(x=features, y=labels, batch_size=2)1features = {
2 "token_ids": np.ones(shape=(2, 12), dtype="int32"),
3 "padding_mask": np.array([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0]] * 2)
4}
5labels = [0, 3]
6
7# Pretrained classifier without preprocessing.
8classifier = keras_hub.models.DistilBertClassifier.from_preset(
9 "hf://keras/distil_bert_base_en",
10 num_classes=4,
11 preprocessor=None,
12)
13classifier.fit(x=features, y=labels, batch_size=2)