NLU error

Describe the bug Following along with the tutorial found at : https://ludwig.ai/latest/examples/nlu/

And using the exact same nlu dataset:

ludwig experiment --dataset nlu.csv --config config.yaml nlu.csv

config.yaml

input_features:
    -
        name: utterance
        type: text
        encoder: 
            type: rnn
            cell_type: lstm
            bidirectional: true
            num_layers: 2
            reduce_output: null
        preprocessing:
            tokenizer: space

output_features:
    -
        name: intent
        type: category
        reduce_input: sum
        decoder:
            num_fc_layers: 1
            output_size: 64
    -
        name: slots
        type: sequence
        decoder: 
            type: tagger

error:

.
.
.
File "/home/martin/.virtualenvs/ludwig/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl
    return forward_call(*input, **kwargs)
  File "/home/martin/.virtualenvs/ludwig/lib/python3.8/site-packages/torch/nn/modules/linear.py", line 114, in forward
    return F.linear(input, self.weight, self.bias)
RuntimeError: mat1 and mat2 shapes cannot be multiplied (2x512 and 256x256)
Training:   0%|   

[justinxzhao]

Hi @martinb-bb, here’s a sample python script you can use:

import logging

import yaml

from ludwig.api import LudwigModel

config = yaml.safe_load("""
input_features:
    -
        name: utterance
        type: text
        encoder: 
            type: rnn
            cell_type: lstm
            bidirectional: true
            num_layers: 2
            reduce_output: null
        preprocessing:
            tokenizer: space

output_features:
    -
        name: intent
        type: category
        reduce_input: sum
        decoder:
            num_fc_layers: 1
            output_size: 64
    -
        name: slots
        type: sequence
        decoder:
            type: tagger

""")

# Define Ludwig model object that drive model training
model = LudwigModel(config=config, logging_level=logging.INFO)

# initiate model training
(
    train_stats,  # dictionary containing training statistics
    preprocessed_data,  # tuple Ludwig Dataset objects of pre-processed training data
    output_directory,  # location of training results stored on disk
) = model.train(dataset="nlu.csv", experiment_name="simple_experiment", model_name="simple_model")

[justinxzhao]

Hi @martinb-bb,

Thanks for filing the issue!

The shape mismatch is an indication that some modules aren’t being initialized properly. I don’t expect this to be a complicated fix, and we should swap this example out with a larger example.

Next quarter, we are planning to back all of our examples and tutorials with real tests that run as part of our CI which should help keep the examples unbroken.

I’ll plan to look at this more closely by next week.