Update init config and recommendations

- As much as I dislike YAML, it seemed like a better format here because it allows us to add comments if we want to explain the different recommendations
- Don't include the generated JS in the repo by default and build it on the fly when running or deploying the site. This ensures it's always up to date.
- Simplify jinja_to_js script and use fewer dependencies

.gitignore

@@ -20,6 +20,7 @@ website/logs
 npm-debug.log*
 website/www/
 website/_deploy.sh
+quickstart-training-generator.js
 
 # Cython / C extensions
 cythonize.json

spacy/cli/init_config.py

@@ -3,17 +3,17 @@ from enum import Enum
 from pathlib import Path
 from wasabi import Printer, diff_strings
 from thinc.api import Config
-from pydantic import BaseModel
 import srsly
 import re
 
 from .. import util
+from ..schemas import RecommendationSchema
 from ._util import init_cli, Arg, Opt, show_validation_error, COMMAND
 
 
-TEMPLATE_ROOT = Path(__file__).parent / "templates"
-TEMPLATE_PATH = TEMPLATE_ROOT / "quickstart_training.jinja"
-RECOMMENDATIONS_PATH = TEMPLATE_ROOT / "quickstart_training_recommendations.json"
+ROOT = Path(__file__).parent / "templates"
+TEMPLATE_PATH = ROOT / "quickstart_training.jinja"
+RECOMMENDATIONS = srsly.read_yaml(ROOT / "quickstart_training_recommendations.yml")
 
 
 class Optimizations(str, Enum):
@@ -21,21 +21,6 @@ class Optimizations(str, Enum):
     accuracy = "accuracy"
 
 
-class RecommendationsTrfItem(BaseModel):
-    name: str
-    size_factor: int
-
-
-class RecommendationsTrf(BaseModel):
-    efficiency: RecommendationsTrfItem
-    accuracy: RecommendationsTrfItem
-
-
-class RecommendationSchema(BaseModel):
-    word_vectors: Optional[str] = None
-    transformer: Optional[RecommendationsTrf] = None
-
-
 @init_cli.command("config")
 def init_config_cli(
     # fmt: off
@@ -111,14 +96,11 @@ def init_config(
         from jinja2 import Template
     except ImportError:
         msg.fail("This command requires jinja2", "pip install jinja2", exits=1)
-    recommendations = srsly.read_json(RECOMMENDATIONS_PATH)
-    lang_defaults = util.get_lang_class(lang).Defaults
-    has_letters = lang_defaults.writing_system.get("has_letters", True)
-    # Filter out duplicates since tok2vec and transformer are added by template
-    pipeline = [pipe for pipe in pipeline if pipe not in ("tok2vec", "transformer")]
-    reco = RecommendationSchema(**recommendations.get(lang, {})).dict()
     with TEMPLATE_PATH.open("r") as f:
         template = Template(f.read())
+    # Filter out duplicates since tok2vec and transformer are added by template
+    pipeline = [pipe for pipe in pipeline if pipe not in ("tok2vec", "transformer")]
+    reco = RecommendationSchema(**RECOMMENDATIONS.get(lang, {})).dict()
     variables = {
         "lang": lang,
         "components": pipeline,
@@ -126,7 +108,7 @@ def init_config(
         "hardware": "cpu" if cpu else "gpu",
         "transformer_data": reco["transformer"],
         "word_vectors": reco["word_vectors"],
-        "has_letters": has_letters,
+        "has_letters": reco["has_letters"],
     }
     base_template = template.render(variables).strip()
     # Giving up on getting the newlines right in jinja for now

spacy/cli/templates/quickstart_training_recommendations.json

@@ -1,13 +0,0 @@
-{
-  "en": {
-    "word_vectors": "en_vectors_web_lg",
-    "transformer": {
-      "efficiency": { "name": "roberta-base", "size_factor": 3 },
-      "accuracy": { "name": "roberta-base", "size_factor": 3 }
-    }
-  },
-  "de": {
-    "word_vectors": null,
-    "transformer": null
-  }
-}

spacy/cli/templates/quickstart_training_recommendations.yml

@@ -0,0 +1,103 @@
+# Recommended settings and available resources for each language, if available.
+# Not all languages have recommended word vecotrs or transformers and for some,
+# the recommended transformer for efficiency and accuracy may be the same.
+en:
+  word_vectors: en_vectors_web_lg
+  transformer:
+    efficiency:
+      name: roberta-base
+      size_factor: 3
+    accuracy:
+      name: roberta-base
+      size_factor: 3
+de:
+  word_vectors: null
+  transformer:
+    efficiency:
+      name: bert-base-german-cased
+      size_factor: 3
+    accuracy:
+      name: bert-base-german-cased
+      size_factor: 3
+fr:
+  word_vectors: null
+  transformer:
+    efficiency:
+      name: camembert-base
+      size_factor: 3
+    accuracy:
+      name: camembert-base
+      size_factor: 3
+es:
+  word_vectors: null
+  transformer:
+    efficiency:
+      name: mrm8488/RuPERTa-base
+      size_factor: 3
+    accuracy:
+      name: mrm8488/RuPERTa-base
+      size_factor: 3
+sv:
+  word_vectors: null
+  transformer:
+    efficiency:
+      name: KB/bert-base-swedish-cased
+      size_factor: 3
+    accuracy:
+      name: KB/bert-base-swedish-cased
+      size_factor: 3
+fi:
+  word_vectors: null
+  transformer:
+    efficiency:
+      name: TurkuNLP/bert-base-finnish-cased-v1
+      size_factor: 3
+    accuracy:
+      name: TurkuNLP/bert-base-finnish-cased-v1
+      size_factor: 3
+el:
+  word_vectors: null
+  transformer:
+    efficiency:
+      name: nlpaueb/bert-base-greek-uncased-v1
+      size_factor: 3
+    accuracy:
+      name: nlpaueb/bert-base-greek-uncased-v1
+      size_factor: 3
+tr:
+  word_vectors: null
+  transformer:
+    efficiency:
+      name: dbmdz/bert-base-turkish-cased
+      size_factor: 3
+    accuracy:
+      name: dbmdz/bert-base-turkish-cased
+      size_factor: 3
+zh:
+  word_vectors: null
+  transformer:
+    efficiency:
+      name: bert-base-chinese
+      size_factor: 3
+    accuracy:
+      name: bert-base-chinese
+      size_factor: 3
+  has_letters: false
+ar:
+  word_vectors: null
+  transformer:
+    efficiency:
+      name: asafaya/bert-base-arabic
+      size_factor: 3
+    accuracy:
+      name: asafaya/bert-base-arabic
+      size_factor: 3
+pl:
+  word_vectors: null
+  transformer:
+    efficiency:
+      name: dkleczek/bert-base-polish-cased-v1
+      size_factor: 3
+    accuracy:
+      name: dkleczek/bert-base-polish-cased-v1
+      size_factor: 3

spacy/schemas.py

@@ -311,3 +311,22 @@ class ProjectConfigSchema(BaseModel):
 
     class Config:
         title = "Schema for project configuration file"
+
+
+# Recommendations for init config workflows
+
+
+class RecommendationTrfItem(BaseModel):
+    name: str
+    size_factor: int
+
+
+class RecommendationTrf(BaseModel):
+    efficiency: RecommendationTrfItem
+    accuracy: RecommendationTrfItem
+
+
+class RecommendationSchema(BaseModel):
+    word_vectors: Optional[str] = None
+    transformer: Optional[RecommendationTrf] = None
+    has_letters: bool = True

spacy/tests/test_cli.py

@@ -2,10 +2,9 @@ import pytest
 from spacy.gold import docs_to_json, biluo_tags_from_offsets
 from spacy.gold.converters import iob2docs, conll_ner2docs, conllu2docs
 from spacy.lang.en import English
-from spacy.schemas import ProjectConfigSchema, validate
+from spacy.schemas import ProjectConfigSchema, RecommendationSchema, validate
 from spacy.cli.pretrain import make_docs
-from spacy.cli.init_config import init_config, RECOMMENDATIONS_PATH
-from spacy.cli.init_config import RecommendationSchema
+from spacy.cli.init_config import init_config, RECOMMENDATIONS
 from spacy.cli._util import validate_project_commands, parse_config_overrides
 from spacy.util import get_lang_class
 import srsly
@@ -335,7 +334,5 @@ def test_init_config(lang, pipeline, optimize):
 
 
 def test_model_recommendations():
-    recommendations = srsly.read_json(RECOMMENDATIONS_PATH)
-    for lang, data in recommendations.items():
-        assert get_lang_class(lang)
+    for lang, data in RECOMMENDATIONS.items():
         assert RecommendationSchema(**data)

website/package.json

@@ -53,7 +53,7 @@
         "remark-react": "^5.0.1"
     },
     "scripts": {
-        "build": "npm run python:setup && gatsby build",
+        "build": "npm run python:install && npm run python:setup && gatsby build",
         "dev": "npm run python:setup && gatsby develop",
         "dev:nightly": "BRANCH=nightly.spacy.io npm run dev",
         "lint": "eslint **",

website/setup/jinja_to_js.py

@@ -11,7 +11,8 @@ from os import path
 from io import StringIO
 from jinja2 import Environment, FileSystemLoader, nodes
 from pathlib import Path
-import typer
+import srsly
+import sys
 
 
 OPERANDS = {
@@ -437,7 +438,8 @@ class JinjaToJS(object):
         with self._interpolation():
             with self._python_bool_wrapper(**kwargs):
                 if node.items:
-                    raise ValueError(f"Can't process non-empty dict in epxression: {node}")
+                    err = f"Can't process non-empty dict in expression: {node}"
+                    raise ValueError(err)
                 self.output.write("{}")
 
     def _process_getattr(self, node, **kwargs):
@@ -1232,18 +1234,22 @@ class JinjaToJS(object):
             self.output.write(")")
 
 
-def main(
-    # fmt: off
-    template_path: Path = typer.Argument(..., exists=True, dir_okay=False, help="Path to .jinja file"),
-    output: Path = typer.Argument(None, help="Path to output module (stdout if unset)"),
-    data_path: Path = typer.Option(None, "--data", help="Optional JSON file with additional data to be included as DATA")
-    # fmt: on
-):
-    """Convert a jinja2 template to a JavaScript module."""
+def main(template_path, output=None, data_path=None):
+    """Convert a jinja2 template to a JavaScript module.
+
+    template_path (Path): Path to .jijna file.
+    output (Optional[Path]): Path to output .js module (stdout if unset).
+    data_path (Optional[Path]): Optional JSON or YAML file with additional data
+        to be included in the JS module as the exported variable DATA.
+    """
     data = "{}"
     if data_path is not None:
-        with data_path.open("r", encoding="utf8") as f:
-            data = json.dumps(json.loads(f.read()))  # dump and load for compactness
+        if data_path.suffix in (".yml", ".yaml"):
+            data = srsly.read_yaml(data_path)
+        else:
+            data = srsly.read_json(data_path)
+        data = srsly.json_dumps(data)  # dump and load for compactness
+    template_path = Path(template_path)
     tpl_file = template_path.parts[-1]
     compiler = JinjaToJS(template_path.parent, tpl_file, js_module_format="es6")
     header = f"// This file was auto-generated by {__file__} based on {tpl_file}"
@@ -1258,4 +1264,10 @@ def main(
 
 
 if __name__ == "__main__":
-    typer.run(main)
+    args = sys.argv[1:]
+    if not len(args):
+        raise ValueError("Need at least one argument: path to .jinja template")
+    template_path = Path(args[0])
+    output = Path(args[1]) if len(args) > 1 else None
+    data_path = Path(args[2]) if len(args) > 2 else None
+    main(template_path, output, data_path)

website/setup/requirements.txt

@@ -1,3 +1,3 @@
 # These are used to compile the training quickstart config
 jinja2
-typer
+srsly

website/setup/setup.sh

@@ -1 +1 @@
-python jinja_to_js.py ../../spacy/cli/templates/quickstart_training.jinja ../src/widgets/quickstart-training-generator.js --data ../../spacy/cli/templates/quickstart_training_recommendations.json
+python jinja_to_js.py ../../spacy/cli/templates/quickstart_training.jinja ../src/widgets/quickstart-training-generator.js ../../spacy/cli/templates/quickstart_training_recommendations.yml

website/src/widgets/quickstart-training.js

@@ -4,7 +4,7 @@ import highlightCode from 'gatsby-remark-prismjs/highlight-code.js'
 
 import { Quickstart } from '../components/quickstart'
 import generator, { DATA as GENERATOR_DATA } from './quickstart-training-generator'
-import { isString, htmlToReact } from '../components/util'
+import { htmlToReact } from '../components/util'
 
 const DEFAULT_LANG = 'en'
 const DEFAULT_HARDWARE = 'gpu'
@@ -47,13 +47,6 @@ const DATA = [
     },
 ]
 
-function stringify(value) {
-    if (isString(value) && value.startsWith('${')) return value
-    const string = JSON.stringify(value)
-    if (Array.isArray(value)) return string.replace(/,/g, ', ')
-    return string
-}
-
 export default function QuickstartTraining({ id, title, download = 'config.cfg' }) {
     const [lang, setLang] = useState(DEFAULT_LANG)
     const [components, setComponents] = useState([])
@@ -73,6 +66,7 @@ export default function QuickstartTraining({ id, title, download = 'config.cfg'
         hardware,
         transformer_data: reco.transformer,
         word_vectors: reco.word_vectors,
+        has_letters: reco.has_letters,
     })
     const rawStr = content.trim().replace(/\n\n\n+/g, '\n\n')
     const rawContent = `${COMMENT}\n${rawStr}`
@@ -90,7 +84,7 @@ export default function QuickstartTraining({ id, title, download = 'config.cfg'
                         id: code,
                         title: name,
                     }))
-                    .sort((a, b) => a.id.localeCompare(b.id))
+                    .sort((a, b) => a.title.localeCompare(b.title))
                 return (
                     <Quickstart
                         download={download}