entity_ruler.patterns fails with srsly.read_jsonl.v1 in spaCy 3.8

[jmyerston]

This following config works in spaCy 3.7 but fails in 3.8:

[initialize.components.entity_ruler.patterns]
@readers = "srsly.read_jsonl.v1"
path = "data/patterns/entity_ruler_patterns.jsonl"

Error:

entity_ruler -> patterns
Input should be an instance of Sequence
{'patterns': <generator object read_jsonl at ...>}

It looks like srsly.read_jsonl.v1 returns a generator, but patterns now requires a Sequence.

Is this an intended change in 3.8?

[aasimansari1]

This is a known breaking change in spaCy 3.8. The srsly.read_jsonl.v1 reader was deprecated and the config path syntax changed.

Fix

Replace the config block with the new spacy.read_labels.v1 reader or load patterns directly in Python:

Option 1 — Load patterns in Python (recommended):

import spacy
import srsly

nlp = spacy.load("your_model")
ruler = nlp.get_pipe("entity_ruler")
patterns = list(srsly.read_jsonl("data/patterns/entity_ruler_patterns.jsonl"))
ruler.add_patterns(patterns)

Option 2 — Update your config.cfg:

[initialize.components.entity_ruler]
patterns_path = "data/patterns/entity_ruler_patterns.jsonl"

The direct patterns_path key was introduced in spaCy 3.6+ and is the preferred way to load patterns at initialisation.

Option 3 — Pin srsly version (temporary workaround):

pip install srsly==2.4.8

See the spaCy 3.8 migration guide for other breaking changes. Hope this unblocks you!