search

tensorflow / text

Making text a first-class citizen in TensorFlow.
https://www.tensorflow.org/beta/tutorials/tensorflow_text/intro
Apache License 2.0
1.22k stars 337 forks source link

Transformers Notebook does not work on Google Colab #1264

Open drewm8080 opened 6 months ago

drewm8080 commented 6 months ago

Gets stuck on this block of code

`embed_pt = PositionalEmbedding(vocab_size=tokenizers.pt.get_vocab_size(), d_model=512) embed_en = PositionalEmbedding(vocab_size=tokenizers.en.get_vocab_size(), d_model=512)

pt_emb = embed_pt(pt) en_emb = embed_en(en`

drewm8080 commented 6 months ago

For this code:

https://www.tensorflow.org/text/tutorials/transformer

MatteoManzari commented 6 months ago

Hi, adding .numpy() after the get_vocab_size() functions will solve your issue!

embed_pt = PositionalEmbedding(vocab_size=tokenizers.pt.get_vocab_size().numpy(), d_model=512) embed_en = PositionalEmbedding(vocab_size=tokenizers.en.get_vocab_size().numpy(), d_model=512)

jemmyfebryan commented 6 months ago

I got error in the importing parts:

AttributeError Traceback (most recent call last) in <cell line: 26>() 24 25 import tensorflow as tf ---> 26 import tensorflow_hub as hub 27 import tensorflow_text as text 28 import tensorflow_datasets as tfds

13 frames /usr/local/lib/python3.10/dist-packages/tf_keras/src/saving/legacy/saved_model/load_context.py in 66 67 ---> 68 tf.internal.register_load_context_function(in_load_context) 69

AttributeError: module 'tensorflow._api.v2.compat.v2.internal' has no attribute 'register_load_context_function'

JuanVargas commented 5 months ago

I got the same problem on May 1, 2024. I am struggling to understand why these tutorials are not tested before releasing code updates. IMHO these frequent issues create a sense of frustration and loss of faith in the tool.

garricklw commented 4 months ago

Adding .numpy() worked for me, posting in the hopes that the notebook can be corrected and the next person can save themselves the hour I wasted trying to fix myself.