Add fittable #140
Add fittable #140
Conversation
Codecov ReportAttention: Patch coverage is
|
| """Save the model to a folder.""" | ||
| save_pipeline(self, path) | ||
|
|
||
| def push_to_hub(self, repo_id: str, token: str | None = None, private: bool = False) -> None: |
There was a problem hiding this comment.
I would add a modelcard and perhaps tags or a library reference, this helps a lot with visibility, usability and findability.
https://huggingface.co/docs/hub/model-cards#specifying-a-library
There was a problem hiding this comment.
This actually already happens because we push the underlying static model to the hub, which has a model card. This model card template is specified in the root of the code.
| self.head = head | ||
|
|
||
| @classmethod | ||
| def from_pretrained( |
There was a problem hiding this comment.
can't we load it from the Hub? perhaps we should align the arguments a bit with the transformers naming given you've also adopted from_pretrained?
For example using pretrained_model_name_or_path. https://huggingface.co/docs/transformers/v4.48.0/en/model_doc/auto#transformers.AutoTokenizer.from_pretrained
There was a problem hiding this comment.
from_pretrained loads from the hub. The arguments mimic the ones from StaticModel and, although they don't match transformers exactly, we're wary of introducing breaking changes.
|
|
||
| # Usage | ||
|
|
||
| Let's assume you're using our `potion-edu classifier`. |
There was a problem hiding this comment.
| Let's assume you're using our `potion-edu classifier`. | |
| Let's assume you're using our [potion-edu classifier](https://huggingface.co/minishlab/potion-8m-edu-classifier). |
Also small todo, don't forget to make this model public
| ```python | ||
| from model2vec.inference import StaticModelPipeline | ||
|
|
||
| s = StaticModelPipeline.from_pretrained("minishlab/potion-8m-edu-classifier") |
There was a problem hiding this comment.
| s = StaticModelPipeline.from_pretrained("minishlab/potion-8m-edu-classifier") | |
| classifier = StaticModelPipeline.from_pretrained("minishlab/potion-8m-edu-classifier") |
| from model2vec.inference import StaticModelPipeline | ||
|
|
||
| s = StaticModelPipeline.from_pretrained("minishlab/potion-8m-edu-classifier") | ||
| label = s.predict("Attitudes towards cattle in the Alps: a study in letting go.") |
There was a problem hiding this comment.
| label = s.predict("Attitudes towards cattle in the Alps: a study in letting go.") | |
| label = classifier.predict("Attitudes towards cattle in the Alps: a study in letting go.") |
|
|
||
| @classmethod | ||
| def from_pretrained( | ||
| cls: type[StaticModelPipeline], path: PathLike, token: str | None = None |
There was a problem hiding this comment.
Should also accept trust_remote_code, and pass it to _load_pipeline I think
|
|
||
| def _predict_and_coerce_to_2d(self, X: list[str] | str) -> np.ndarray: | ||
| """Predict the labels of the input and coerce the output to a matrix.""" | ||
| encoded = self.model.encode(X) |
There was a problem hiding this comment.
There's no control over encode since this is in a private function, I'm not sure if we want to give any control here? E.g batch size, multiprocessing, etc?
|
|
||
| def configure_optimizers(self) -> OptimizerLRScheduler: | ||
| """Simple Adam optimizer.""" | ||
| optimizer = torch.optim.Adam(self.model.parameters(), lr=1e-3) |
There was a problem hiding this comment.
I think learning rate is now ignored, this should be:
| optimizer = torch.optim.Adam(self.model.parameters(), lr=1e-3) | |
| optimizer = torch.optim.Adam(self.model.parameters(), lr=self.learning_rate) |
| @@ -0,0 +1,248 @@ | |||
| { | |||
There was a problem hiding this comment.
This prints in the notebook are not saved/pushed I think. This is a bit weird, for example when you have:
"Pretty good! We outperform the tf-idf pipeline by a wide margin.", it doesn't actually show it in the notebook (see https://github.com/MinishLab/model2vec/blob/add-fittable/tutorials/train_classifier.ipynb).
No description provided.