Incompatibility of spacy models as ray reference

[suneeta-mall]

How to reproduce the behaviour

When running spacy with ray, for inference usecase, an error ValueError: buffer source array is read-only is encountered. Full stack is shown below:

  File "~/.venv/lib/python3.8/site-packages/ray/_private/worker.py", line 2380, in get
    raise value.as_instanceof_cause()
ray.exceptions.RayTaskError(ValueError): ray::generate() (pid=596931, ip=172.16.147.156)
  File "test.py", line 9, in generate
    doc = list(nlp.pipe([corpus]))[0]
  File "~/.venv/lib/python3.8/site-packages/spacy/language.py", line 1618, in pipe
    for doc in docs:
  File "~/.venv/lib/python3.8/site-packages/spacy/util.py", line 1703, in _pipe
    yield from proc.pipe(docs, **kwargs)
  File "spacy/pipeline/transition_parser.pyx", line 245, in pipe
  File "~/.venv/lib/python3.8/site-packages/spacy/util.py", line 1650, in minibatch
    batch = list(itertools.islice(items, int(batch_size)))
  File "~/.venv/lib/python3.8/site-packages/spacy/util.py", line 1703, in _pipe
    yield from proc.pipe(docs, **kwargs)
  File "spacy/pipeline/transition_parser.pyx", line 245, in pipe
  File "~/.venv/lib/python3.8/site-packages/spacy/util.py", line 1650, in minibatch
    batch = list(itertools.islice(items, int(batch_size)))
  File "~/.venv/lib/python3.8/site-packages/spacy/util.py", line 1703, in _pipe
    yield from proc.pipe(docs, **kwargs)
  File "spacy/pipeline/pipe.pyx", line 55, in pipe
  File "~/.venv/lib/python3.8/site-packages/spacy/util.py", line 1703, in _pipe
    yield from proc.pipe(docs, **kwargs)
  File "spacy/pipeline/pipe.pyx", line 55, in pipe
  File "~/.venv/lib/python3.8/site-packages/spacy/util.py", line 1703, in _pipe
    yield from proc.pipe(docs, **kwargs)
  File "spacy/pipeline/trainable_pipe.pyx", line 73, in pipe
  File "~/.venv/lib/python3.8/site-packages/spacy/util.py", line 1650, in minibatch
    batch = list(itertools.islice(items, int(batch_size)))
  File "~/.venv/lib/python3.8/site-packages/spacy/util.py", line 1703, in _pipe
    yield from proc.pipe(docs, **kwargs)
  File "spacy/pipeline/trainable_pipe.pyx", line 79, in pipe
  File "~/.venv/lib/python3.8/site-packages/spacy/util.py", line 1722, in raise_error
    raise e
  File "spacy/pipeline/trainable_pipe.pyx", line 75, in spacy.pipeline.trainable_pipe.TrainablePipe.pipe
  File "~/.venv/lib/python3.8/site-packages/spacy/pipeline/tok2vec.py", line 126, in predict
    tokvecs = self.model.predict(docs)
  File "~/.venv/lib/python3.8/site-packages/thinc/model.py", line 334, in predict
    return self._func(self, X, is_train=False)[0]
  File "~/.venv/lib/python3.8/site-packages/thinc/layers/chain.py", line 54, in forward
    Y, inc_layer_grad = layer(X, is_train=is_train)
  File "~/.venv/lib/python3.8/site-packages/thinc/model.py", line 310, in __call__
    return self._func(self, X, is_train=is_train)
  File "~/.venv/lib/python3.8/site-packages/thinc/layers/chain.py", line 54, in forward
    Y, inc_layer_grad = layer(X, is_train=is_train)
  File "~/.venv/lib/python3.8/site-packages/thinc/model.py", line 310, in __call__
    return self._func(self, X, is_train=is_train)
  File "~/.venv/lib/python3.8/site-packages/thinc/layers/with_array.py", line 36, in forward
    return cast(Tuple[SeqT, Callable], _ragged_forward(model, Xseq, is_train))
  File "~/.venv/lib/python3.8/site-packages/thinc/layers/with_array.py", line 91, in _ragged_forward
    Y, get_dX = layer(Xr.dataXd, is_train)
  File "~/.venv/lib/python3.8/site-packages/thinc/model.py", line 310, in __call__
    return self._func(self, X, is_train=is_train)
  File "~/.venv/lib/python3.8/site-packages/thinc/layers/concatenate.py", line 57, in forward
    Ys, callbacks = zip(*[layer(X, is_train=is_train) for layer in model.layers])
  File "~/.venv/lib/python3.8/site-packages/thinc/layers/concatenate.py", line 57, in <listcomp>
    Ys, callbacks = zip(*[layer(X, is_train=is_train) for layer in model.layers])
  File "~/.venv/lib/python3.8/site-packages/thinc/model.py", line 310, in __call__
    return self._func(self, X, is_train=is_train)
  File "~/.venv/lib/python3.8/site-packages/thinc/layers/chain.py", line 54, in forward
    Y, inc_layer_grad = layer(X, is_train=is_train)
  File "~/.venv/lib/python3.8/site-packages/thinc/model.py", line 310, in __call__
    return self._func(self, X, is_train=is_train)
  File "~/.venv/lib/python3.8/site-packages/thinc/layers/hashembed.py", line 72, in forward
    output = model.ops.gather_add(vectors, keys)
  File "thinc/backends/numpy_ops.pyx", line 460, in thinc.backends.numpy_ops.NumpyOps.gather_add
  File "stringsource", line 660, in View.MemoryView.memoryview_cwrapper
  File "stringsource", line 350, in View.MemoryView.memoryview.__cinit__
ValueError: buffer source array is read-only

Code to reproduce is:

from typing import List

import ray
import spacy


@ray.remote
def generate(nlp, corpus: str) -> List[str]:
    doc = list(nlp.pipe([corpus]))[0]
    return doc.noun_chunks


if __name__ == "__main__":
    ray.init()
    nlp = spacy.load(name="en_core_sci_sm")
    nlp_ref = ray.put(nlp)
    texts = ["sfdfdl?", "dgfhgfhjgj"] 
    ref_ids = [generate.remote(nlp_ref, text) for text in texts]

    while len(ref_ids):
        processed, unprocessed = ray.wait(ref_ids)
        ref_ids = unprocessed
        if processed:
            print(ray.get(processed))

Note: The error disappears when the model is initialised in action instead of as a ray reference.

I expected it to just work?

Your Environment

• Operating System: Linux/mac
• Python Version Used: 3.8.10
• spaCy Version Used: '3.7.4'
• Environment Information: