haystack/test/test_modeling_prediction_head.py
Sara Zan a59bca3661
Apply black formatting (#2115)
* Testing black on ui/

* Applying black on docstores

* Add latest docstring and tutorial changes

* Create a single GH action for Black and docs to reduce commit noise to the minimum, slightly refactor the OpenAPI action too

* Remove comments

* Relax constraints on pydoc-markdown

* Split temporary black from the docs. Pydoc-markdown was obsolete and needs a separate PR to upgrade

* Fix a couple of bugs

* Add a type: ignore that was missing somehow

* Give path to black

* Apply Black

* Apply Black

* Relocate a couple of type: ignore

* Update documentation

* Make Linux CI run after applying Black

* Triggering Black

* Apply Black

* Remove dependency, does not work well

* Remove manually double trailing commas

* Update documentation

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
2022-02-03 13:43:18 +01:00

31 lines
993 B
Python

import logging
from haystack.modeling.model.adaptive_model import AdaptiveModel
from haystack.modeling.model.language_model import LanguageModel
from haystack.modeling.model.prediction_head import QuestionAnsweringHead
from haystack.modeling.utils import set_all_seeds, initialize_device_settings
def test_prediction_head_load_save(tmp_path, caplog=None):
if caplog:
caplog.set_level(logging.CRITICAL)
set_all_seeds(seed=42)
devices, n_gpu = initialize_device_settings(use_cuda=False)
lang_model = "bert-base-german-cased"
language_model = LanguageModel.load(lang_model)
prediction_head = QuestionAnsweringHead()
model = AdaptiveModel(
language_model=language_model,
prediction_heads=[prediction_head],
embeds_dropout_prob=0.1,
lm_output_types=["per_sequence"],
device=devices[0],
)
model.save(tmp_path)
model_loaded = AdaptiveModel.load(tmp_path, device="cpu")
assert model_loaded is not None