Automatically download converted models from the Hugging Face Hub (#70)

* Automatically download converted models from the Hugging Face Hub

* Remove unused import

* Remove non needed requirements in dev mode

* Remove extra index URL when pip install in CI

* Allow downloading to a specific directory

* Update docstring

* Add argument to disable the progess bars

* Fix typo in docstring

tests/conftest.py

@@ -1,6 +1,5 @@
 import os
 
-import ctranslate2
 import pytest
 
 
@@ -12,20 +11,3 @@ def data_dir():
 @pytest.fixture
 def jfk_path(data_dir):
     return os.path.join(data_dir, "jfk.flac")
-
-
-@pytest.fixture(scope="session")
-def tiny_model_dir(tmp_path_factory):
-    model_path = str(tmp_path_factory.mktemp("data") / "model")
-    convert_model("tiny", model_path)
-    return model_path
-
-
-def convert_model(size, output_dir):
-    name = "openai/whisper-%s" % size
-
-    ctranslate2.converters.TransformersConverter(
-        name,
-        copy_files=["tokenizer.json"],
-        load_as_float16=True,
-    ).convert(output_dir, quantization="float16")

tests/test_transcribe.py

@@ -1,8 +1,8 @@
 from faster_whisper import WhisperModel
 
 
-def test_transcribe(tiny_model_dir, jfk_path):
-    model = WhisperModel(tiny_model_dir)
+def test_transcribe(jfk_path):
+    model = WhisperModel("tiny")
     segments, info = model.transcribe(jfk_path, word_timestamps=True)
 
     assert info.language == "en"

tests/test_utils.py

@@ -0,0 +1,17 @@
+import os
+
+from faster_whisper import download_model
+
+
+def test_download_model(tmpdir):
+    output_dir = str(tmpdir.join("model"))
+
+    model_dir = download_model("tiny", output_dir=output_dir)
+
+    assert model_dir == output_dir
+    assert os.path.isdir(model_dir)
+    assert not os.path.islink(model_dir)
+
+    for filename in os.listdir(model_dir):
+        path = os.path.join(model_dir, filename)
+        assert not os.path.islink(path)