TorchBiLSTMEncoder RuntimeError due to dimension mismatch

[vistamou]

Hello,

I'm trying to use BiLSTM archictecture: [components.tok2vec.model.encode]
@architectures = "spacy.TorchBiLSTMEncoder.v1"
width = 64
depth = 2
dropout = ${training.dropout}

but I get the following error when I try to train the model: .../nn/modules/rnn.py", line 203, in check_input
expected_input_dim, input.dim()))
RuntimeError: input must have 3 dimensions, got 2

could you provide some clue why?

[polm]

Can you please share the rest of your config and the versions of spaCy and related libraries? Hard to say what's going on without more information...

[vistamou]

Yes, sorry here it is the config file:

[paths]
train = null
dev = null
vectors = null
init_tok2vec = null

[system]
gpu_allocator = null
seed = 0

[nlp]
lang = "en"
pipeline = ["tok2vec","tagger"]
before_creation = null
after_creation = null
after_pipeline_creation = null
batch_size = 256


[components]

[components.tok2vec]
factory = "tok2vec"

[components.tok2vec.model]
@architectures = "spacy.Tok2Vec.v1"

[components.tok2vec.model.embed]
@architectures = "spacy.MultiHashEmbed.v1"
width = ${components.tok2vec.model.encode.width}
attrs = ["ORTH", "SHAPE"]
rows = [5000, 2500]
include_static_vectors = True

[components.tok2vec.model.encode]
@architectures = "spacy.TorchBiLSTMEncoder.v1"
width = 64
depth = 2
dropout = ${training.dropout}

[components.tagger]
factory = "tagger"
overwrite = false
#scorer = {"@scorers":"spacy.tagger_scorer.v1"}

[components.tagger.model]
@architectures = "spacy.Tagger.v1"
nO = null

[components.tagger.model.tok2vec]
@architectures = "spacy.Tok2VecListener.v1"
width = ${components.tok2vec.model.encode:width}
#upstream = "tok2vec"


[corpora]

[corpora.dev]
@readers = "spacy.Corpus.v1"
path = ${paths.dev}
gold_preproc = false
max_length = 0
limit = 0
augmenter = null

[corpora.train]
@readers = "spacy.Corpus.v1"
path = ${paths.train}
gold_preproc = false
max_length = 0
limit = 0
augmenter = null

[training]
train_corpus = "corpora.train"
dev_corpus = "corpora.dev"
seed = ${system:seed}
gpu_allocator = ${system:gpu_allocator}
dropout = 0.1
accumulate_gradient = 1
patience = 5000
max_epochs = 0
max_steps = 0
eval_frequency = 1000
frozen_components = []
before_to_disk = null
annotating_components = []

[training.batcher]
@batchers = "spacy.batch_by_words.v1"
discard_oversize = false
tolerance = 0.2
get_length = null

[training.batcher.size]
@schedules = "compounding.v1"
start = 100
stop = 1000
compound = 1.001
t = 0.0

[training.logger]
@loggers = "spacy.ConsoleLogger.v1"
progress_bar = false

[training.optimizer]
@optimizers = "Adam.v1"
beta1 = 0.9
beta2 = 0.999
L2_is_weight_decay = true
L2 = 0.01
grad_clip = 1.0
use_averages = true
eps = 0.00000001
learn_rate = 0.001

[training.score_weights]
tag_acc = 0.32
dep_uas = 0.0
dep_las = 0.32
dep_las_per_type = null
sents_p = null
sents_r = null
sents_f = 0.04
ents_f = 0.32
ents_p = 0.0
ents_r = 0.0
ents_per_type = null
speed = 0.0

[pretraining]

[initialize]
vocab_data = null
vectors = ${paths.vectors}
init_tok2vec = ${paths.init_tok2vec}
before_init = null
after_init = null

[initialize.components]

[initialize.components.parser]

[initialize.components.parser.labels]
@readers = "spacy.read_labels.v1"
path = "corpus/labels/parser.json"
require = false

[initialize.components.tagger]

[initialize.components.tagger.labels]
@readers = "spacy.read_labels.v1"
path = "corpus/labels/tagger.json"
require = false

[initialize.lookups]
@misc = "spacy.LookupsDataLoader.v1"
lang = ${nlp.lang}
tables = []

spacy version and related : spacy 3.2.0
spacy-alignments 0.8.4
spacy-legacy 3.0.8
spacy-loggers 1.0.1
spacy-lookups-data 1.0.3
spacy-ray 0.1.4
spacy-transformers 1.1.2

Thank you!

[polm]

Thanks for the extra details - I was able to reproduce this error. This might be a bug, we'll look at it more.

[polm]

Hey, it looks like this is a bug in Tok2Vec.v1 that was fixed for Tok2Vec.v2 - see #6442. If you update that line in your config (under [components.tok2vec.model]) to use v2 this should just work.

How did you generate this config?

[vistamou]

it works, thanks a lot :)))

[polm]

Glad to hear it works!

If you could let us know how you generated your config, that'd be a big help. v2 was released a while ago, but if we have examples using v1 lying around we would want to update them.