Feature Extraction
Transformers
Safetensors
jolia
medical
radiology
ct
3d
vision
foundation-model
self-supervised
custom_code
Instructions to use raidium/Jolia with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use raidium/Jolia with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("feature-extraction", model="raidium/Jolia", trust_remote_code=True)# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("raidium/Jolia", trust_remote_code=True, dtype="auto") - Notebooks
- Google Colab
- Kaggle
| """Minimal stand-ins for the ``raidium.rd.hub`` base classes. | |
| The vendored Atlas modules (``jolia_multimodal_atlas`` etc.) were copied | |
| verbatim from the internal ``raidium.rd.models`` library, which subclasses | |
| ``raidium.rd.hub.utils.{config.Config, base_model.BaseModel, | |
| base_preprocessing.BasePreprocessing}``. Those base classes carry a lot of | |
| hub-registry machinery the public Hugging Face release does not need, so we | |
| replace them with the three tiny shims below. Behaviour relevant to inference | |
| (config field defaults, ``self.config`` storage, ``nn.Module`` semantics) is | |
| preserved exactly. | |
| """ | |
| from __future__ import annotations | |
| import torch.nn as nn | |
| class Config: | |
| """Lightweight replacement for ``raidium.rd.hub.utils.config.Config``. | |
| Sets every field annotated anywhere in the MRO from the matching keyword | |
| argument, falling back to the class-level default. This reproduces the | |
| pydantic-style ``MyConfig(field=value)`` construction the vendored Atlas | |
| configs rely on, without the pydantic dependency. | |
| """ | |
| def __init__(self, **kwargs: object) -> None: | |
| fields: dict[str, object] = {} | |
| for klass in reversed(type(self).__mro__): | |
| fields.update(getattr(klass, "__annotations__", {})) | |
| for name in fields: | |
| setattr(self, name, kwargs.get(name, getattr(type(self), name, None))) | |
| class BaseModel(nn.Module): | |
| """Replacement for ``raidium.rd.hub.utils.base_model.BaseModel``. | |
| Only the constructor contract used by the vendored backbone is kept: store | |
| ``config`` and behave as a plain ``nn.Module``. Hub push/pull methods are | |
| intentionally absent — Hugging Face's ``PreTrainedModel`` owns that role. | |
| """ | |
| config_class = None | |
| def __init__(self, config: Config) -> None: | |
| super().__init__() | |
| self.config = config | |
| class BasePreprocessing(nn.Module): | |
| """Replacement for ``raidium.rd.hub.utils.base_preprocessing.BasePreprocessing``.""" | |
| config_class = None | |
| transform_input = "metadata" | |
| def __init__(self, config: Config) -> None: | |
| super().__init__() | |
| self.config = config | |