From 351bedff30a99f740044740412558032b642a9ee Mon Sep 17 00:00:00 2001 From: Stanislas Laurent Date: Sun, 5 Jul 2026 20:24:01 +0200 Subject: [PATCH 01/24] feat(data): add InputModality.text + text metadata inference (refs #340) --- src/raitap/data/metadata.py | 2 ++ src/raitap/data/tests/test_text_metadata.py | 17 +++++++++++++++++ src/raitap/data/tests/test_types.py | 2 +- src/raitap/data/types.py | 2 ++ 4 files changed, 22 insertions(+), 1 deletion(-) create mode 100644 src/raitap/data/tests/test_text_metadata.py diff --git a/src/raitap/data/metadata.py b/src/raitap/data/metadata.py index da2e89b0..36a18ddc 100644 --- a/src/raitap/data/metadata.py +++ b/src/raitap/data/metadata.py @@ -47,6 +47,8 @@ def infer_data_input_metadata(config: object, data: object) -> DataInputMetadata return DataInputMetadata(kind="image", shape=shape, layout="NCHW") if modality is InputModality.tabular: return DataInputMetadata(kind="tabular", shape=shape, layout="(B,F)") + if modality is InputModality.text: + return DataInputMetadata(kind="text", shape=shape, layout="TOKENS") # Fallback for callers with no recorded modality (e.g. direct/test use): # sniff the source path as before. diff --git a/src/raitap/data/tests/test_text_metadata.py b/src/raitap/data/tests/test_text_metadata.py new file mode 100644 index 00000000..3c9a2f69 --- /dev/null +++ b/src/raitap/data/tests/test_text_metadata.py @@ -0,0 +1,17 @@ +import torch + +from raitap.data.metadata import infer_data_input_metadata +from raitap.data.types import InputModality + + +class _Data: + input_metadata = None + tensor = torch.zeros((2, 8), dtype=torch.long) + input_modality = InputModality.text + source = "" + + +def test_text_modality_infers_text_kind_and_tokens_layout() -> None: + md = infer_data_input_metadata(config=object(), data=_Data()) + assert md.kind == "text" + assert md.layout == "TOKENS" diff --git a/src/raitap/data/tests/test_types.py b/src/raitap/data/tests/test_types.py index 1ccf640d..d6884ab6 100644 --- a/src/raitap/data/tests/test_types.py +++ b/src/raitap/data/tests/test_types.py @@ -2,7 +2,7 @@ def test_input_modality_has_only_todays_members() -> None: - assert {m.value for m in InputModality} == {"image", "tabular"} + assert {m.value for m in InputModality} == {"image", "tabular", "text"} def test_modality_extensions_cover_every_member() -> None: diff --git a/src/raitap/data/types.py b/src/raitap/data/types.py index 943f28d9..6c5c8414 100644 --- a/src/raitap/data/types.py +++ b/src/raitap/data/types.py @@ -55,6 +55,7 @@ class InputModality(StrEnum): image = "image" tabular = "tabular" + text = "text" #: Single source of truth for which file extensions map to which modality. @@ -63,4 +64,5 @@ class InputModality(StrEnum): MODALITY_EXTENSIONS: dict[InputModality, frozenset[str]] = { InputModality.image: frozenset({".jpg", ".jpeg", ".png", ".bmp", ".webp"}), InputModality.tabular: frozenset({".csv", ".tsv", ".parquet"}), + InputModality.text: frozenset({".txt"}), } From 572e99cc8348558f71d9b9d9a72113bf12bb2a6c Mon Sep 17 00:00:00 2001 From: Stanislas Laurent Date: Sun, 5 Jul 2026 20:41:19 +0200 Subject: [PATCH 02/24] feat(model): load HuggingFace text-classification model + tokenizer (refs #340) --- pyproject.toml | 5 + src/raitap/configs/schema.py | 2 + src/raitap/models/model.py | 8 +- .../models/tests/test_hf_text_backend.py | 19 ++ src/raitap/models/torch_backend.py | 27 +- uv.lock | 250 +++++++++++++++++- 6 files changed, 305 insertions(+), 6 deletions(-) create mode 100644 src/raitap/models/tests/test_hf_text_backend.py diff --git a/pyproject.toml b/pyproject.toml index 8849f23b..6dd022f5 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -140,6 +140,11 @@ transparency = ["raitap[shap,captum]"] # this extra stays hardware-variant-free. xgboost = ["xgboost>=2.0", "scikit-learn>=1.3", "torch>=2.8.0,<2.9.0"] +# Text model backend (HuggingFace) — AutoModelForSequenceClassification + +# AutoTokenizer loading for the text input modality. Does not bundle torch: +# relies on whatever torch-* extra is already installed for the pipeline. +text = ["transformers>=4.40"] + # Robustness module torchattacks = ["torchattacks>=3.5.1"] foolbox = ["foolbox>=3.3.4", "numba>=0.59"] diff --git a/src/raitap/configs/schema.py b/src/raitap/configs/schema.py index 8f8e75d4..8820e5b5 100644 --- a/src/raitap/configs/schema.py +++ b/src/raitap/configs/schema.py @@ -66,6 +66,8 @@ class ModelConfig: # the TaskKind value (e.g. "detection") for custom models the inference # can't recognise. task_kind: TaskKind | None = None + # HuggingFace tokenizer id/path for text models. Set => text modality. + tokenizer: str | None = None @dataclass diff --git a/src/raitap/models/model.py b/src/raitap/models/model.py index e0761470..8da06691 100644 --- a/src/raitap/models/model.py +++ b/src/raitap/models/model.py @@ -20,7 +20,7 @@ from . import xgboost_backend # noqa: F401 # pyright: ignore[reportUnusedImport] from .onnx_backend import OnnxBackend from .runtime import resolve_torch_device -from .torch_backend import TorchBackend +from .torch_backend import TorchBackend, load_hf_text_backend if TYPE_CHECKING: import torch @@ -81,6 +81,12 @@ def _load_model(self, config: AppConfig, *, allow_unsafe_pickle: bool = False) - " model.source: resnet50 (built-in demo model)" ) + tokenizer = getattr(config.model, "tokenizer", None) + if tokenizer: + return load_hf_text_backend( + source, tokenizer=tokenizer, device=resolve_torch_device(hardware) + ) + path = Path(source) suffix = path.suffix.lower() diff --git a/src/raitap/models/tests/test_hf_text_backend.py b/src/raitap/models/tests/test_hf_text_backend.py new file mode 100644 index 00000000..a3313527 --- /dev/null +++ b/src/raitap/models/tests/test_hf_text_backend.py @@ -0,0 +1,19 @@ +import pytest + +pytest.importorskip("transformers") + +import torch + +from raitap.models.torch_backend import load_hf_text_backend + + +def test_hf_text_backend_forwards_ids_and_mask() -> None: + backend = load_hf_text_backend( + "hf-internal-testing/tiny-random-DistilBertForSequenceClassification", + tokenizer="hf-internal-testing/tiny-random-DistilBertForSequenceClassification", + device=torch.device("cpu"), + ) + enc = backend.tokenizer(["hello world", "bye"], padding=True, return_tensors="pt") + out = backend(enc["input_ids"], attention_mask=enc["attention_mask"]) + logits = out.logits if hasattr(out, "logits") else out + assert logits.shape[0] == 2 diff --git a/src/raitap/models/torch_backend.py b/src/raitap/models/torch_backend.py index e62b70c3..58d1afb7 100644 --- a/src/raitap/models/torch_backend.py +++ b/src/raitap/models/torch_backend.py @@ -51,6 +51,10 @@ def __init__( self.device = torch.device("cpu") if device is None else device self._task_kind = task_kind if task_kind is not None else self._infer_task_kind(model) self.category_names = category_names + # ``load_hf_text_backend`` sets this to an ``AutoTokenizer`` instance; + # declared here so pyright sees the attribute on every ``TorchBackend`` + # (harmless ``None`` otherwise). + self.tokenizer: Any = None @classmethod def from_path( @@ -105,7 +109,11 @@ def prepare_detection_inputs(self, inputs: list[torch.Tensor]) -> list[torch.Ten def _prepare_kwargs(self, kwargs: dict[str, Any]) -> dict[str, Any]: return _move_tensors_to_device(kwargs, self.device) - def __call__(self, inputs: torch.Tensor | list[torch.Tensor]) -> Any: + def __call__(self, inputs: torch.Tensor | list[torch.Tensor], **kwargs: Any) -> Any: + # ``**kwargs`` covers text models' ``attention_mask`` (and similar HF + # keyword-only forward args); image/tabular callers never pass any. + if kwargs: + return self.model(inputs, **self._prepare_kwargs(kwargs)) return self.model(inputs) def autograd_module(self) -> nn.Module: @@ -132,3 +140,20 @@ def _torch_hardware_label(device: torch.device) -> str: "mps": "Apple MPS", } return label_by_type.get(device.type, device.type.upper()) + + +def load_hf_text_backend(source: str, *, tokenizer: str, device: torch.device) -> TorchBackend: + """Load a HuggingFace sequence-classification model + tokenizer. + + ``source`` and ``tokenizer`` are HuggingFace hub ids or local directories, + resolved via ``AutoModelForSequenceClassification``/``AutoTokenizer``. + Returns a ``TorchBackend`` with ``.tokenizer`` set. + """ + # lazy: `transformers` is an optional dep (extra "text"), imported only + # when a text model is actually requested. + from transformers import AutoModelForSequenceClassification, AutoTokenizer + + module = AutoModelForSequenceClassification.from_pretrained(source).to(device).eval() + backend = TorchBackend(module, device=device, task_kind=TaskKind.classification) + backend.tokenizer = AutoTokenizer.from_pretrained(tokenizer) + return backend diff --git a/uv.lock b/uv.lock index f19040da..20e47b9d 100644 --- a/uv.lock +++ b/uv.lock @@ -598,14 +598,14 @@ wheels = [ [[package]] name = "click" -version = "8.3.2" +version = "8.4.2" source = { registry = "https://pypi.org/simple" } dependencies = [ { name = "colorama", marker = "sys_platform == 'win32' or (extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-onnx-intel') or (extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-torch-intel') or (extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-cuda') or (extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-intel') or (extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-cuda') or (extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-intel') or (extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-onnx-intel') or (extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-cpu') or (extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-intel') or (extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-xgboost') or (extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cpu') or (extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cuda') or (extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-xgboost') or (extra == 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-cuda') or (extra == 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-intel') or (extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-torch-intel') or (extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-xgboost') or (extra == 'extra-6-raitap-torch-intel' and extra == 'extra-6-raitap-xgboost')" }, ] -sdist = { url = "https://files.pythonhosted.org/packages/57/75/31212c6bf2503fdf920d87fee5d7a86a2e3bcf444984126f13d8e4016804/click-8.3.2.tar.gz", hash = "sha256:14162b8b3b3550a7d479eafa77dfd3c38d9dc8951f6f69c78913a8f9a7540fd5", size = 302856, upload-time = "2026-04-03T19:14:45.118Z" } +sdist = { url = "https://files.pythonhosted.org/packages/76/d4/81420972a676e8ffea40450d8c8c92943e7218a78fe9b64359836cc9876b/click-8.4.2.tar.gz", hash = "sha256:9a6cea6e60b17ebe0a44c5cc636d94f09bd66142c1cd7d8b4cd731c4917a15f6", size = 338000, upload-time = "2026-06-24T17:45:15.148Z" } wheels = [ - { url = "https://files.pythonhosted.org/packages/e4/20/71885d8b97d4f3dde17b1fdb92dbd4908b00541c5a3379787137285f602e/click-8.3.2-py3-none-any.whl", hash = "sha256:1924d2c27c5653561cd2cae4548d1406039cb79b858b747cfea24924bbc1616d", size = 108379, upload-time = "2026-04-03T19:14:43.505Z" }, + { url = "https://files.pythonhosted.org/packages/fb/e2/79c688af8b210d232694e31e59da9f6ec747bae31c3f5946e4e9b98860d5/click-8.4.2-py3-none-any.whl", hash = "sha256:e6f9f66136c816745b9d65817da91d61d957fb16e02e4dcd0552553c5a197b76", size = 119243, upload-time = "2026-06-24T17:45:13.73Z" }, ] [[package]] @@ -1319,6 +1319,58 @@ wheels = [ { url = "https://files.pythonhosted.org/packages/04/4b/29cac41a4d98d144bf5f6d33995617b185d14b22401f75ca86f384e87ff1/h11-0.16.0-py3-none-any.whl", hash = "sha256:63cf8bbe7522de3bf65932fda1d9c2772064ffb3dae62d55932da54b31cb6c86", size = 37515, upload-time = "2025-04-24T03:35:24.344Z" }, ] +[[package]] +name = "hf-xet" +version = "1.5.1" +source = { registry = "https://pypi.org/simple" } +sdist = { url = "https://files.pythonhosted.org/packages/4b/2d/57fd21d84d93efb4bd0b962383790e19dd1bc053501b4264c97903b4e83e/hf_xet-1.5.1.tar.gz", hash = "sha256:51ef4500dab3764b41135ee1381a4b62ce56fc54d4c92b719b59e597d6df5bf6", size = 876636, upload-time = "2026-06-08T23:02:53.897Z" } +wheels = [ + { url = "https://files.pythonhosted.org/packages/64/ee/dd9ba7beae1005e54131b7d45263cc74c8a066d47d354e6d58ae9445a388/hf_xet-1.5.1-cp313-cp313t-macosx_10_12_x86_64.whl", hash = "sha256:dbf48c0d02cf0b2e568944330c60d9120c272dabe013bd892d48e25bc6797577", size = 4069485, upload-time = "2026-06-08T23:02:13.193Z" }, + { url = "https://files.pythonhosted.org/packages/b6/bc/9cae6cfeb4e03070874e73e5c97c66eb90369d3206b6a2b1ef5f96520888/hf_xet-1.5.1-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:e78e4e5192ad2b674c2e1160b651cb9134db974f8ae1835bdfbfb0166b894a43", size = 3838493, upload-time = "2026-06-08T23:02:15.282Z" }, + { url = "https://files.pythonhosted.org/packages/ba/b4/d5c01e0eb6d9f2ca2dacd84d0d1b71e6cfbb2ef3208c968528e010e9b3d7/hf_xet-1.5.1-cp313-cp313t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:6f7a04a8ad962422e225bc49fbbac99dc1806764b1f3e54dbd154bffa7593947", size = 4505658, upload-time = "2026-06-08T23:02:17.196Z" }, + { url = "https://files.pythonhosted.org/packages/76/c5/29a7598c0c6383c523dc22186d577f4e04267a626cd95ae60f67c00bfe66/hf_xet-1.5.1-cp313-cp313t-manylinux_2_28_aarch64.whl", hash = "sha256:d48199c2bf4f8df0adc55d31d1368b6ec0e4d4f45bc86b08038089c23db0bed8", size = 4292822, upload-time = "2026-06-08T23:02:18.608Z" }, + { url = "https://files.pythonhosted.org/packages/04/9a/dceaf6ca69390126b86ea825fb354b93d01163199070b7bd849225de9468/hf_xet-1.5.1-cp313-cp313t-musllinux_1_2_aarch64.whl", hash = "sha256:97f212a88d14bbf573619a74b7fecb238de77d08fc702e54dec6f78276ca3283", size = 4491255, upload-time = "2026-06-08T23:02:20.124Z" }, + { url = "https://files.pythonhosted.org/packages/48/a7/e5a7afaacf6c1791fdbeeac42951fb81c3d2bc482992b115dedcc86d963e/hf_xet-1.5.1-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:f61e3665892a6c8c5e765395838b8ddf36185da835253d4bc4509a81e49fb342", size = 4711062, upload-time = "2026-06-08T23:02:21.863Z" }, + { url = "https://files.pythonhosted.org/packages/53/49/2802f8433c9742ce281bddc1e65c02c32268ca3098d66828b05e12e45ee2/hf_xet-1.5.1-cp313-cp313t-win_amd64.whl", hash = "sha256:f4ad3ebd4c32dd2b27099d69dc7b2df821e30767e46fb6ee6a0713778243b8ff", size = 4017205, upload-time = "2026-06-08T23:02:23.495Z" }, + { url = "https://files.pythonhosted.org/packages/9e/5a/50c71195b9fb883659f596e7252faf4c18c58e753a9013bdbf9bac5d2250/hf_xet-1.5.1-cp313-cp313t-win_arm64.whl", hash = "sha256:8298485c1e36e7e67cbd01eeb1376619b7af43d4f1ec245caae306f890a8a32d", size = 3845426, upload-time = "2026-06-08T23:02:25.124Z" }, + { url = "https://files.pythonhosted.org/packages/7a/d8/5e54cf37434759d1f4f2ba9b66077ff9d4c4e1f37b6bd7975da5c40d94ab/hf_xet-1.5.1-cp37-abi3-macosx_10_12_x86_64.whl", hash = "sha256:6abd35c3221eff63836618ddfb954dcf84798603f71d8e33e3ed7b04acfdbe6e", size = 4077794, upload-time = "2026-06-08T23:02:40.656Z" }, + { url = "https://files.pythonhosted.org/packages/35/94/4b2ecfbad8f8b04701a23aefb62f540b9137d058b7e1dbef16a32676f0e9/hf_xet-1.5.1-cp37-abi3-macosx_11_0_arm64.whl", hash = "sha256:94e761bbd266bf4c03cee73753916062665ce8365aa40ed321f45afcb934b41e", size = 3845354, upload-time = "2026-06-08T23:02:42.702Z" }, + { url = "https://files.pythonhosted.org/packages/de/cc/f99f4bc7295023d7bd9ebbfd51f75cc530ca262c1227666268b8208f4b77/hf_xet-1.5.1-cp37-abi3-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:892e3a3a3aecc12aded8b93cf4f9cd059282c7de0732f7d55026f3abdf474350", size = 4514864, upload-time = "2026-06-08T23:02:44.497Z" }, + { url = "https://files.pythonhosted.org/packages/cd/6e/21f7e5a2381278bd3b7b7a5a4d90038518bb6308a0c1daf5d9f8268bb178/hf_xet-1.5.1-cp37-abi3-manylinux_2_28_aarch64.whl", hash = "sha256:a93df2039190502835b1db8cd7e178b0b7b889fe9ab51299d5ced26e0dd879a4", size = 4303784, upload-time = "2026-06-08T23:02:46.203Z" }, + { url = "https://files.pythonhosted.org/packages/35/0e/f992bb6927ac1cb30ef74e62268f551f338bc32b2191f7c96a44c6f7283e/hf_xet-1.5.1-cp37-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:0c97106032ef70467b4f6bc2d0ccc266d7613ee076afc56516c502f87ce1c4a6", size = 4500703, upload-time = "2026-06-08T23:02:47.628Z" }, + { url = "https://files.pythonhosted.org/packages/fb/d1/90a498d05447980b977b1669246eeeeae4cfb0ea3e7a286eaba627f91bf9/hf_xet-1.5.1-cp37-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:6208adb15d192b90e4c2ad2a27ed864359b2cb0f2494eb6d7c7f3699ac02e2bf", size = 4719498, upload-time = "2026-06-08T23:02:49.268Z" }, + { url = "https://files.pythonhosted.org/packages/6d/b6/20f99cfe97cc663a711f7b33cc21d4793e51968e9a26125b4afcd77315ba/hf_xet-1.5.1-cp37-abi3-win_amd64.whl", hash = "sha256:f7b3002f95d1c13e24bcb4537baa8f0eb3838957067c91bb4959bc004a6435f5", size = 4026419, upload-time = "2026-06-08T23:02:50.829Z" }, + { url = "https://files.pythonhosted.org/packages/f9/fa/77453694888f03e5a8c8852d1514a0894d8e81c622d39edbaf308ea0dcf4/hf_xet-1.5.1-cp37-abi3-win_arm64.whl", hash = "sha256:93d090b57b211133f6c0dab0205ef5cb6d89162979ba75a74845045cc3063b8e", size = 3855178, upload-time = "2026-06-08T23:02:52.452Z" }, +] + +[[package]] +name = "httpcore" +version = "1.0.9" +source = { registry = "https://pypi.org/simple" } +dependencies = [ + { name = "certifi" }, + { name = "h11" }, +] +sdist = { url = "https://files.pythonhosted.org/packages/06/94/82699a10bca87a5556c9c59b5963f2d039dbd239f25bc2a63907a05a14cb/httpcore-1.0.9.tar.gz", hash = "sha256:6e34463af53fd2ab5d807f399a9b45ea31c3dfa2276f15a2c3f00afff6e176e8", size = 85484, upload-time = "2025-04-24T22:06:22.219Z" } +wheels = [ + { url = "https://files.pythonhosted.org/packages/7e/f5/f66802a942d491edb555dd61e3a9961140fd64c90bce1eafd741609d334d/httpcore-1.0.9-py3-none-any.whl", hash = "sha256:2d400746a40668fc9dec9810239072b40b4484b640a8c38fd654a024c7a1bf55", size = 78784, upload-time = "2025-04-24T22:06:20.566Z" }, +] + +[[package]] +name = "httpx" +version = "0.28.1" +source = { registry = "https://pypi.org/simple" } +dependencies = [ + { name = "anyio" }, + { name = "certifi" }, + { name = "httpcore" }, + { name = "idna" }, +] +sdist = { url = "https://files.pythonhosted.org/packages/b1/df/48c586a5fe32a0f01324ee087459e112ebb7224f646c0b5023f5e79e9956/httpx-0.28.1.tar.gz", hash = "sha256:75e98c5f16b0f35b567856f597f06ff2270a374470a5c2392242528e3e3e42fc", size = 141406, upload-time = "2024-12-06T15:37:23.222Z" } +wheels = [ + { url = "https://files.pythonhosted.org/packages/2a/39/e50c7c3a983047577ee07d2a9e53faf5a69493943ec3f6a384bdc792deb2/httpx-0.28.1-py3-none-any.whl", hash = "sha256:d909fcccc110f8c7faf814ca82a9a4d816bc5a6dbfea25d6591d6985b8ba59ad", size = 73517, upload-time = "2024-12-06T15:37:21.509Z" }, +] + [[package]] name = "huey" version = "2.6.0" @@ -1328,6 +1380,26 @@ wheels = [ { url = "https://files.pythonhosted.org/packages/1a/34/fae9ac8f1c3a552fd3f7ff652b94c78d219dedc5fce0c0a4232457760a00/huey-2.6.0-py3-none-any.whl", hash = "sha256:1b9df9d370b49c6d5721ba8a01ac9a787cf86b3bdc584e4679de27b920395c3f", size = 76951, upload-time = "2026-01-06T03:01:00.808Z" }, ] +[[package]] +name = "huggingface-hub" +version = "1.22.0" +source = { registry = "https://pypi.org/simple" } +dependencies = [ + { name = "click" }, + { name = "filelock" }, + { name = "fsspec" }, + { name = "hf-xet", marker = "platform_machine == 'AMD64' or platform_machine == 'aarch64' or platform_machine == 'amd64' or platform_machine == 'arm64' or platform_machine == 'x86_64' or (extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-onnx-intel') or (extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-torch-intel') or (extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-cuda') or (extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-intel') or (extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-cuda') or (extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-intel') or (extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-onnx-intel') or (extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-cpu') or (extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-intel') or (extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-xgboost') or (extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cpu') or (extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cuda') or (extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-xgboost') or (extra == 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-cuda') or (extra == 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-intel') or (extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-torch-intel') or (extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-xgboost') or (extra == 'extra-6-raitap-torch-intel' and extra == 'extra-6-raitap-xgboost')" }, + { name = "httpx" }, + { name = "packaging" }, + { name = "pyyaml" }, + { name = "tqdm" }, + { name = "typing-extensions" }, +] +sdist = { url = "https://files.pythonhosted.org/packages/77/ea/dc54b4dda5841cb3a7812a178695be776e7c15c597887c2ed892f17d015a/huggingface_hub-1.22.0.tar.gz", hash = "sha256:e2dfe5fe1ec3b87ba2709aa34555b23e3f3f6ad4d7255238e13ddb8348e6bbfa", size = 914232, upload-time = "2026-07-03T09:46:44.685Z" } +wheels = [ + { url = "https://files.pythonhosted.org/packages/64/9c/a1a377265abd8b823a2c661c665028ccb6b9fba1ca9d08e52ff679c20ecd/huggingface_hub-1.22.0-py3-none-any.whl", hash = "sha256:b09e19309ae09ee0a71892701c4fe70af39ab4e00817321dc62f2289a977249b", size = 765085, upload-time = "2026-07-03T09:46:42.832Z" }, +] + [[package]] name = "hydra-core" version = "1.3.2" @@ -3472,6 +3544,9 @@ shap = [ { name = "shap", version = "0.50.0", source = { registry = "https://pypi.org/simple" }, marker = "(platform_machine == 'x86_64' and sys_platform == 'darwin') or (platform_machine != 'x86_64' and extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine != 'x86_64' and extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-torch-intel') or (platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-cuda') or (platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-cuda') or (platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-intel') or (platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-cpu') or (platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-intel') or (platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-xgboost') or (platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cpu') or (platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cuda') or (platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-xgboost') or (platform_machine != 'x86_64' and extra == 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-cuda') or (platform_machine != 'x86_64' and extra == 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-intel') or (platform_machine != 'x86_64' and extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-torch-intel') or (platform_machine != 'x86_64' and extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-xgboost') or (platform_machine != 'x86_64' and extra == 'extra-6-raitap-torch-intel' and extra == 'extra-6-raitap-xgboost') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-onnx-intel') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-cuda') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-intel') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-cuda') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-intel') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-onnx-intel') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-cpu') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-intel') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-xgboost') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cpu') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cuda') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-xgboost') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-cuda') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-intel') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-torch-intel') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-xgboost') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-torch-intel' and extra == 'extra-6-raitap-xgboost') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-auto-lirpa' and extra != 'extra-6-raitap-torch-cpu' and extra != 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-torch-intel')" }, { name = "shap", version = "0.51.0", source = { registry = "https://pypi.org/simple" }, marker = "platform_machine != 'x86_64' or sys_platform != 'darwin' or (extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-onnx-intel') or (extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-torch-intel') or (extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-cuda') or (extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-intel') or (extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-cuda') or (extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-intel') or (extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-onnx-intel') or (extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-cpu') or (extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-intel') or (extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-xgboost') or (extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cpu') or (extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cuda') or (extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-xgboost') or (extra == 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-cuda') or (extra == 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-intel') or (extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-torch-intel') or (extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-xgboost') or (extra == 'extra-6-raitap-torch-intel' and extra == 'extra-6-raitap-xgboost')" }, ] +text = [ + { name = "transformers" }, +] torch-cpu = [ { name = "torch", version = "2.8.0", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "(platform_machine != 's390x' and platform_machine != 'x86_64' and sys_platform == 'darwin' and extra == 'extra-6-raitap-torch-cpu') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-cuda') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-cpu') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-intel') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-xgboost') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cpu') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cuda') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-xgboost') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-auto-lirpa' and extra != 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-intel') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cpu' and extra != 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-cuda') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cpu' and extra != 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-intel') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra != 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-torch-intel') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra != 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-xgboost') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra != 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-intel' and extra == 'extra-6-raitap-xgboost') or (platform_machine == 's390x' and extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine == 's390x' and extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-torch-intel') or (platform_machine == 's390x' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-cuda') or (platform_machine == 's390x' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine == 's390x' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-cuda') or (platform_machine == 's390x' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-intel') or (platform_machine == 's390x' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine == 's390x' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-cpu') or (platform_machine == 's390x' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-intel') or (platform_machine == 's390x' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-xgboost') or (platform_machine == 's390x' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cpu') or (platform_machine == 's390x' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cuda') or (platform_machine == 's390x' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-xgboost') or (platform_machine == 's390x' and extra == 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-cuda') or (platform_machine == 's390x' and extra == 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-intel') or (platform_machine == 's390x' and extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-torch-intel') or (platform_machine == 's390x' and extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-xgboost') or (platform_machine == 's390x' and extra == 'extra-6-raitap-torch-intel' and extra == 'extra-6-raitap-xgboost') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-torch-intel') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-cuda') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-cuda') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-intel') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-cpu') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-intel') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-xgboost') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cpu') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cuda') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-xgboost') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-cuda') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-intel') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-torch-intel') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-xgboost') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-torch-intel' and extra == 'extra-6-raitap-xgboost') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-onnx-intel') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-cuda') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-intel') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-cuda') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-intel') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-onnx-intel') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-cpu') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-intel') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-xgboost') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cpu') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cuda') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-xgboost') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-cuda') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-intel') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-torch-intel') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-xgboost') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-torch-intel' and extra == 'extra-6-raitap-xgboost') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-auto-lirpa' and extra != 'extra-6-raitap-torch-cpu' and extra != 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-torch-intel')" }, { name = "torch", version = "2.8.0+cpu", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "(platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-torch-intel') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-cuda') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-cuda') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-intel') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-cpu') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-intel') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-xgboost') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cpu') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cuda') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-xgboost') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-cuda') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-intel') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-torch-intel') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-xgboost') or (platform_machine != 's390x' and platform_machine != 'x86_64' and extra == 'extra-6-raitap-torch-intel' and extra == 'extra-6-raitap-xgboost') or (platform_machine == 's390x' and extra == 'extra-6-raitap-torch-cpu') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-torch-cpu') or (platform_machine == 's390x' and extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine == 's390x' and extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-torch-intel') or (platform_machine == 's390x' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-cuda') or (platform_machine == 's390x' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine == 's390x' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-cuda') or (platform_machine == 's390x' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-intel') or (platform_machine == 's390x' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine == 's390x' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-intel') or (platform_machine == 's390x' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-xgboost') or (platform_machine == 's390x' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cuda') or (platform_machine == 's390x' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-xgboost') or (platform_machine == 's390x' and extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-torch-intel') or (platform_machine == 's390x' and extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-xgboost') or (platform_machine == 's390x' and extra == 'extra-6-raitap-torch-intel' and extra == 'extra-6-raitap-xgboost') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-torch-intel') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-cuda') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-cuda') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-intel') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-onnx-intel') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-intel') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-xgboost') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cuda') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-xgboost') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-torch-intel') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-xgboost') or (platform_machine == 'x86_64' and extra == 'extra-6-raitap-torch-intel' and extra == 'extra-6-raitap-xgboost') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-torch-cpu') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-onnx-intel') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-cuda') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-intel') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-cuda') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-intel') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-onnx-intel') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-intel') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-xgboost') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cuda') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-xgboost') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-torch-intel') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-xgboost') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-torch-intel' and extra == 'extra-6-raitap-xgboost') or (sys_platform != 'darwin' and extra == 'extra-6-raitap-auto-lirpa' and extra != 'extra-6-raitap-torch-cpu' and extra != 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-torch-intel')" }, @@ -3604,11 +3679,12 @@ requires-dist = [ { name = "torchvision", marker = "extra == 'onnx-cuda'", specifier = ">=0.20.0,<0.24.0", index = "https://download.pytorch.org/whl/cu126", conflict = { package = "raitap", extra = "onnx-cuda" } }, { name = "torchvision", marker = "extra == 'torch-cpu'", specifier = ">=0.20.0,<0.24.0", index = "https://download.pytorch.org/whl/cpu", conflict = { package = "raitap", extra = "torch-cpu" } }, { name = "torchvision", marker = "extra == 'torch-cuda'", specifier = ">=0.20.0,<0.24.0", index = "https://download.pytorch.org/whl/cu126", conflict = { package = "raitap", extra = "torch-cuda" } }, + { name = "transformers", marker = "extra == 'text'", specifier = ">=4.40" }, { name = "triton-xpu", marker = "python_full_version < '3.14' and sys_platform != 'darwin' and extra == 'onnx-intel'", specifier = ">=3.0.0", index = "https://download.pytorch.org/whl/xpu", conflict = { package = "raitap", extra = "onnx-intel" } }, { name = "triton-xpu", marker = "python_full_version < '3.14' and sys_platform != 'darwin' and extra == 'torch-intel'", specifier = ">=3.0.0", index = "https://download.pytorch.org/whl/xpu", conflict = { package = "raitap", extra = "torch-intel" } }, { name = "xgboost", marker = "extra == 'xgboost'", specifier = ">=2.0" }, ] -provides-extras = ["torch-cpu", "torch-cuda", "torch-intel", "onnx-cpu", "onnx-cuda", "onnx-intel", "shap", "captum", "quantus", "transparency", "xgboost", "torchattacks", "foolbox", "imagecorruptions", "marabou", "auto-lirpa", "robustness", "launcher", "mlflow", "tracking", "metrics", "pdf", "html", "reporting"] +provides-extras = ["torch-cpu", "torch-cuda", "torch-intel", "onnx-cpu", "onnx-cuda", "onnx-intel", "shap", "captum", "quantus", "transparency", "xgboost", "text", "torchattacks", "foolbox", "imagecorruptions", "marabou", "auto-lirpa", "robustness", "launcher", "mlflow", "tracking", "metrics", "pdf", "html", "reporting"] [package.metadata.requires-dev] dev = [ @@ -3654,6 +3730,78 @@ tooling = [ { name = "pre-commit", specifier = ">=4.0.0" }, ] +[[package]] +name = "regex" +version = "2026.6.28" +source = { registry = "https://pypi.org/simple" } +sdist = { url = "https://files.pythonhosted.org/packages/f1/05/e4f219230e11e774a6c9987d2ab0d0c6b8573e13a17e143d0015bee710ef/regex-2026.6.28.tar.gz", hash = "sha256:3cb4b6c5cb3060cc31efdc1fbb27c25fb9b29044afd87e40601a1c4d9db54342", size = 416101, upload-time = "2026-06-28T19:56:55.302Z" } +wheels = [ + { url = "https://files.pythonhosted.org/packages/72/db/9051b36294bdbabaa9c7db57db0fbcdfbd17f7a106c539bb423d0323faea/regex-2026.6.28-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:a71b51dd08b9b62f055fafab3dee8af8bd2ec81b373a44caef18d6c5ca28f43a", size = 489481, upload-time = "2026-06-28T19:53:36.684Z" }, + { url = "https://files.pythonhosted.org/packages/35/3f/24097a3c3ff30f9a639888900faaecabcf5f54a5bc9c851c297e11b349ef/regex-2026.6.28-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:9c26a47770d30a0f85c01e261d2a3ebc342c4af6fd666dbd8c1fe4cbf3adf726", size = 291292, upload-time = "2026-06-28T19:53:38.39Z" }, + { url = "https://files.pythonhosted.org/packages/5e/cc/e0d762a189cfb4e8926d16e691720690d139a977b38fdb80230c259332ab/regex-2026.6.28-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:e5efbc1af38f97e300d43028e5a92e752d924bcfb7f465d8669d5d5a6e78c233", size = 289232, upload-time = "2026-06-28T19:53:40.181Z" }, + { url = "https://files.pythonhosted.org/packages/4b/c8/ca0ac7f09cc88ca61e0c61c53f7db29334f660ffba5d0b52378e7c44723c/regex-2026.6.28-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:f1758df6fdd8c800620a5638958720e8a635e1da49a2f09df2dd63e94a24ec4a", size = 792332, upload-time = "2026-06-28T19:53:41.782Z" }, + { url = "https://files.pythonhosted.org/packages/8e/92/04ae94cbe0dd1f478b2aef6c46f995bb6946d3e338d4b28605478b66a2b7/regex-2026.6.28-cp311-cp311-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:ad73ecf20c1ef5c975639f8bf845a9370fcf7dada7edc1e3b0bca20e2f8202f6", size = 861743, upload-time = "2026-06-28T19:53:43.261Z" }, + { url = "https://files.pythonhosted.org/packages/4c/ec/024d7638c807679ff8a0e6081d01d66c7762339af1cac71e45911587ff9a/regex-2026.6.28-cp311-cp311-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:4d80c798b0eec6ea3d45f8816a1e8886c5664615d347d89e8c075b576a1b5a5d", size = 906481, upload-time = "2026-06-28T19:53:44.948Z" }, + { url = "https://files.pythonhosted.org/packages/cd/fd/93bfe5af45f0be4fa8983945455c0e6924e1aeb879cde227958869c1e71c/regex-2026.6.28-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:a361feeaf1b6ba1df060f2ff5c5947092edf537a35ce78e76387ac56d3e0f4a4", size = 799867, upload-time = "2026-06-28T19:53:46.997Z" }, + { url = "https://files.pythonhosted.org/packages/ee/fd/e5d965d41f2398c8ce0f37a4652f03bb297fd009bb796d390134225dda12/regex-2026.6.28-cp311-cp311-manylinux_2_31_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:8b92366d9c8bba9642989534073662abdd9b41faf7603a7ae71597833f3b88f0", size = 773632, upload-time = "2026-06-28T19:53:48.892Z" }, + { url = "https://files.pythonhosted.org/packages/eb/d9/ff39afaec92b9ee2dba0302a4783976005091681069808938c31cf8df3b6/regex-2026.6.28-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:11251768cc23f097dd61b18f67966e70f74da822784d17e12a444eb6b29d4288", size = 781669, upload-time = "2026-06-28T19:53:50.693Z" }, + { url = "https://files.pythonhosted.org/packages/45/4e/e2fd4bb8228e10c24af2d7ff867182372190e498eab9fd29cbe54c403c95/regex-2026.6.28-cp311-cp311-musllinux_1_2_ppc64le.whl", hash = "sha256:ad5c67786145ec28a71a267d9f9d92bdc8d70d65541eea852c253f520a01f918", size = 854497, upload-time = "2026-06-28T19:53:52.323Z" }, + { url = "https://files.pythonhosted.org/packages/72/7c/f0340384a973082979064156d05f3d2cc1dced7371efcd7a1b45726a1a8a/regex-2026.6.28-cp311-cp311-musllinux_1_2_riscv64.whl", hash = "sha256:f1da438e739765c3e85175ede05816cbede3caaacb1e0680568bda6119bfdfca", size = 763335, upload-time = "2026-06-28T19:53:54.024Z" }, + { url = "https://files.pythonhosted.org/packages/e1/32/90ce0d0898e205506cc22b9c81cfb16b722e06ca5f50fad51c053c2a727b/regex-2026.6.28-cp311-cp311-musllinux_1_2_s390x.whl", hash = "sha256:d98b639046e51c5de64d9f77351532105e99ca271cb6f7640e1f903d6ab63032", size = 844615, upload-time = "2026-06-28T19:53:56.216Z" }, + { url = "https://files.pythonhosted.org/packages/6a/ef/55abb149599dce1ade687170557129524011eeb3d92afe02429cea7754a2/regex-2026.6.28-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:1e164ace4dbab5c6ad4a4ac7c41a2638fe226d0c770a86f2eb041f594bac6ee7", size = 789193, upload-time = "2026-06-28T19:53:57.791Z" }, + { url = "https://files.pythonhosted.org/packages/f8/ea/cf7f6f6f152e52fdad978b913bf24c14df647eca0f81ef31f3aee0be8982/regex-2026.6.28-cp311-cp311-win32.whl", hash = "sha256:3169a3159e4d99d9ae85ff0ed90ef3b8906cc3152653b6078b842ace6c8f72c3", size = 266731, upload-time = "2026-06-28T19:53:59.938Z" }, + { url = "https://files.pythonhosted.org/packages/c6/cf/a48d8e8d406b22481cad146f48fa0dfca3c5f402b91f26d8e5a0fe4f513d/regex-2026.6.28-cp311-cp311-win_amd64.whl", hash = "sha256:5977295b0a74e8241df8a4b3b27b12412a831f6fa32ee8b755039592cd768c3d", size = 277918, upload-time = "2026-06-28T19:54:01.502Z" }, + { url = "https://files.pythonhosted.org/packages/89/b2/a222392207db7ed86281a732a99f7cf7f2bb35d332799e892b8510be000e/regex-2026.6.28-cp311-cp311-win_arm64.whl", hash = "sha256:f5fbaef40c3e9282ccee4b075f5600a0d858aa0c34147732f1baa69c8188a95d", size = 276876, upload-time = "2026-06-28T19:54:03.411Z" }, + { url = "https://files.pythonhosted.org/packages/da/21/44aa415873032056c43eac21c67285deb2cf66cddb2a964c3cdc8f803efc/regex-2026.6.28-cp312-cp312-macosx_10_13_universal2.whl", hash = "sha256:81cc5793ad33a10444445e8d29d3c73e752c8fb2e120772d70fcb6d41df40fe1", size = 490480, upload-time = "2026-06-28T19:54:05.392Z" }, + { url = "https://files.pythonhosted.org/packages/8b/5f/30d4116093c2128099f78b6990dfc1698fdbf3ee528f1e1c647378034c79/regex-2026.6.28-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:e18225243250a1f7d7e5e5d883f3b96465cd79031acf5c6db902b7025f2125d9", size = 292137, upload-time = "2026-06-28T19:54:07.088Z" }, + { url = "https://files.pythonhosted.org/packages/cb/0e/ca20a0e0de49837e6337603a91ab77556aa27033ac5b975615d98698cfb3/regex-2026.6.28-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:ecd1638b1c2db1f2d01c182a4b0d3e2e88b0e99910320a745c1727ee3638ddab", size = 289623, upload-time = "2026-06-28T19:54:08.762Z" }, + { url = "https://files.pythonhosted.org/packages/50/11/c013422a7e2c59946df8ac93e792a4922c98287f2a2181341603c78a5d98/regex-2026.6.28-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:4303ebe16b74eeb3fe2715745023266fea92fd44a23f3e7bb2fb48c7a7bbc195", size = 796756, upload-time = "2026-06-28T19:54:10.616Z" }, + { url = "https://files.pythonhosted.org/packages/b0/95/1309645a0e1ee6fb91d954501da57a0b33d50ad2a9acb313702851a7054e/regex-2026.6.28-cp312-cp312-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:56b856b70b96c381d837f609eee442a1bd320cd2159f5c294b679552fb1a7eaf", size = 865465, upload-time = "2026-06-28T19:54:12.742Z" }, + { url = "https://files.pythonhosted.org/packages/20/06/491802db47c6f5e2904ffa2518ad3ac27fe6bbf5a66d73210a95cc080d47/regex-2026.6.28-cp312-cp312-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:f74675ab76ab1d005ffba4dee308e53e89efc22be6e9f9fae5b539a3f81bdff2", size = 912350, upload-time = "2026-06-28T19:54:14.508Z" }, + { url = "https://files.pythonhosted.org/packages/5e/60/3ba57840bcc7e2367090360de0c15a5ba6ad22be89314251105f2e943f43/regex-2026.6.28-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:90581684565a93f7258af1e5d3f41ef20d7d7c61f2a428183a342bcb65485e38", size = 801261, upload-time = "2026-06-28T19:54:16.432Z" }, + { url = "https://files.pythonhosted.org/packages/eb/27/af1eb74e9a78c782b3e450b611a595e44906da8a5107e1227f4a7fd0480b/regex-2026.6.28-cp312-cp312-manylinux_2_31_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:28f9e6c28f9b90f6f784595a33240a57e181e61b6ee3dc259b25c61e356d1aa3", size = 777072, upload-time = "2026-06-28T19:54:18.128Z" }, + { url = "https://files.pythonhosted.org/packages/20/18/fdd4c883a39e3ed00d669062af1135809bfd3281bf528150849fbd68825b/regex-2026.6.28-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:378a71d861fc7c8806b04ac5b133d53c0e774f92f5d9663a539872d3fa2b0417", size = 785119, upload-time = "2026-06-28T19:54:20.314Z" }, + { url = "https://files.pythonhosted.org/packages/1c/79/0aabe34b8482dcadf64355f70f96e22eba5ec6c1efb33563f89654f4061c/regex-2026.6.28-cp312-cp312-musllinux_1_2_ppc64le.whl", hash = "sha256:4cc199874ecd6267a49b111052250825bfe19b5101b23b2ba80f54efa3e0994e", size = 860118, upload-time = "2026-06-28T19:54:22.368Z" }, + { url = "https://files.pythonhosted.org/packages/a8/2c/c973323306a27c9db7d160e9584eb7e0ece2a96224ccb0d39060558b31f9/regex-2026.6.28-cp312-cp312-musllinux_1_2_riscv64.whl", hash = "sha256:b916a10431494ef4b4d62c6c89cab6426af7873125b8cd6c15811bf5fc58eec8", size = 765786, upload-time = "2026-06-28T19:54:24.265Z" }, + { url = "https://files.pythonhosted.org/packages/e3/df/9ca3e378e352242a4cb45573a5e9162c3ee791507702a23966fa559e36b5/regex-2026.6.28-cp312-cp312-musllinux_1_2_s390x.whl", hash = "sha256:2e27727fba075f1e4409416d2f537d4c30fc11f012ea507f7bd74d3e19ecb57a", size = 852120, upload-time = "2026-06-28T19:54:25.972Z" }, + { url = "https://files.pythonhosted.org/packages/a2/3e/3e31e255c4971f53cbce6306b5e3c76cbd3735a54f419bb3b2f194e9f68c/regex-2026.6.28-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:700fc6a7844bb2c4149292ac79d1df8841a00acd4d45cd32c1ebc7bcc1fd0da8", size = 789503, upload-time = "2026-06-28T19:54:27.678Z" }, + { url = "https://files.pythonhosted.org/packages/72/01/d36561c21c3033d7eeb31d51b491916817de7861acefccc5fc9db8a5037c/regex-2026.6.28-cp312-cp312-win32.whl", hash = "sha256:03376d60b6a11aecb88a79fa2be06b40faa01c6693bc31ef69435cd4818b9463", size = 267109, upload-time = "2026-06-28T19:54:29.316Z" }, + { url = "https://files.pythonhosted.org/packages/a0/59/bbbb0591f38b18c65977cd65ce64749eba1c1996c99ac04e900fc30c0dcb/regex-2026.6.28-cp312-cp312-win_amd64.whl", hash = "sha256:fbd2ded482bf99e6651992bbfcde460272724d4bbc49ef3d6b46d9312867ec84", size = 277711, upload-time = "2026-06-28T19:54:31.143Z" }, + { url = "https://files.pythonhosted.org/packages/86/06/be4f6b337d773ae5739a1bc238f97c16926e72017243735853c030f4c628/regex-2026.6.28-cp312-cp312-win_arm64.whl", hash = "sha256:37294d3d7ddb64c7e89184b2894e0f8f0a19c514bc59513d71fe692c3a8d5fc6", size = 277022, upload-time = "2026-06-28T19:54:32.97Z" }, + { url = "https://files.pythonhosted.org/packages/b6/53/d5c1b3cc0b5a0c985563ad6fac93d73ff2b300cb84342d89f044625d6bc7/regex-2026.6.28-cp313-cp313-macosx_10_13_universal2.whl", hash = "sha256:b295a83426e0e44e9e60fde99789e181bd26788a1890ae7fe2a24c69bb6246ca", size = 490329, upload-time = "2026-06-28T19:54:35.775Z" }, + { url = "https://files.pythonhosted.org/packages/8d/9f/0c3503e819e91ca0e7a901a8e989ebf840ac7c7aea20b1fc7f31b6759f77/regex-2026.6.28-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:0c31665c0deb5c111557a1cac8c27bd5629e2f9e7fd5058900a03576c33b601c", size = 292039, upload-time = "2026-06-28T19:54:37.977Z" }, + { url = "https://files.pythonhosted.org/packages/bb/7f/cd004e13fcad23b3794a82307dfd222e6365eb7f598bd3caab148a830bff/regex-2026.6.28-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:6bf295f2c59de77d1ea7de053607ae4dc9ceb3d57bbb6c7ec51ef4acc4ccff94", size = 289488, upload-time = "2026-06-28T19:54:39.545Z" }, + { url = "https://files.pythonhosted.org/packages/73/4c/293fb34586fbcdc47eac436069e9c11f71fae5dadfd4889b475d7d2e5f7a/regex-2026.6.28-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:17c077586770f67e05bbffeba07fbee6b2b22244f4d4caf8d94e59d574befe04", size = 796772, upload-time = "2026-06-28T19:54:41.347Z" }, + { url = "https://files.pythonhosted.org/packages/92/fa/c0cd1a90b7d12d9dc155cfc8bdea8df9720988ea5b07e8fa1eccbd0ab2dd/regex-2026.6.28-cp313-cp313-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:0e6cb5a61486f9062397d2e189573b39d38ecfaed698fd9fb6e2756a8ebb8762", size = 865467, upload-time = "2026-06-28T19:54:43.485Z" }, + { url = "https://files.pythonhosted.org/packages/4e/db/0b479973046d005a1eaea299d5d536aeecb9488a16d9cbb8286338102e2d/regex-2026.6.28-cp313-cp313-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:e86e91a2664f44c3a4e363a7d78fb17c27d5046882e30ea5a877f5e89b28d2ba", size = 912345, upload-time = "2026-06-28T19:54:46.091Z" }, + { url = "https://files.pythonhosted.org/packages/5b/5b/d65adfbd02f32212431bca1f06d1e2eb763a20b12978b454bafaf23dacb7/regex-2026.6.28-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:4dfd1331c49233998d84fc5f1f4436cf7a435a7655f6cf0f490229bb5c7254e5", size = 801291, upload-time = "2026-06-28T19:54:48.3Z" }, + { url = "https://files.pythonhosted.org/packages/fc/09/2103686defaf9a0a31c1663782359d5b45f42524c64cca681f5481e44a5e/regex-2026.6.28-cp313-cp313-manylinux_2_31_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:cadea12805a1bce0b091c302b814207be26fb60a9c0e7f9ad2f9e21790a429fe", size = 777106, upload-time = "2026-06-28T19:54:50.326Z" }, + { url = "https://files.pythonhosted.org/packages/85/5a/b57593c0aa23ed269ec332fbcf07852abcb6b746e811d9464e0d09b4e25f/regex-2026.6.28-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:5f2c1682b67ad5d2376498f2a5a2a8f782fa2e4a06d0465b5e357799806e8a20", size = 785175, upload-time = "2026-06-28T19:54:52.172Z" }, + { url = "https://files.pythonhosted.org/packages/79/59/c36e756ad29bf14d7b6c6d7138952476b21f6160286cedb98ac13481c993/regex-2026.6.28-cp313-cp313-musllinux_1_2_ppc64le.whl", hash = "sha256:64e142eb55e84868087da1375d7c36ff97d55010951849f515322a91d5fef1b4", size = 860186, upload-time = "2026-06-28T19:54:54.11Z" }, + { url = "https://files.pythonhosted.org/packages/61/66/49808aea0da9649c300139360708fb91b7144be1f962fcebf96755fde948/regex-2026.6.28-cp313-cp313-musllinux_1_2_riscv64.whl", hash = "sha256:abb4daabe7be63273787a62dfd6164dadf8f7a63fbec3d2730e5e5e7126d858c", size = 765754, upload-time = "2026-06-28T19:54:56.04Z" }, + { url = "https://files.pythonhosted.org/packages/be/c5/52bbd436cf2200decdf48825fa38363eaaeebb77011ea9928a1ef9e0b9f2/regex-2026.6.28-cp313-cp313-musllinux_1_2_s390x.whl", hash = "sha256:ec2b2ad00ab8c16a2798cc8db80c53c4d5b8b3a2441f6cbaef06625f5ca25854", size = 852085, upload-time = "2026-06-28T19:54:57.988Z" }, + { url = "https://files.pythonhosted.org/packages/f9/c3/0390b66e3019497143fe768b3ba567b64d8b24f3812d09506deb86f4a0f0/regex-2026.6.28-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:bfc9677982c914d9085b8e1c3b3ae6e88f139fb56531c2416d6c8f338093c22b", size = 789600, upload-time = "2026-06-28T19:54:59.977Z" }, + { url = "https://files.pythonhosted.org/packages/88/fd/ab5b03653a244975069fed93d73f4f5f7484c03a84cedb238292510d7182/regex-2026.6.28-cp313-cp313-win32.whl", hash = "sha256:bf54bc693fc4e0530e666ba5ec4bcba14dbe8f66b7cfc15c27317d1a6e40b9a5", size = 267088, upload-time = "2026-06-28T19:55:02.159Z" }, + { url = "https://files.pythonhosted.org/packages/68/55/21022f7d3143210ae8d4ff905c45306237b657375cc0b97883f49db3d423/regex-2026.6.28-cp313-cp313-win_amd64.whl", hash = "sha256:e128feaf65bf3d9eb91bec92322a8f7e4835e9c798f3e9ea4b69f4def85620e3", size = 277680, upload-time = "2026-06-28T19:55:04.185Z" }, + { url = "https://files.pythonhosted.org/packages/b6/99/7f664804f1aef924542b0b233996b78b3e4d0a52d9951358aac99f129f51/regex-2026.6.28-cp313-cp313-win_arm64.whl", hash = "sha256:695873e0ea8d3815ea9e92e2c68faf039cc450e2c0a62a31afe2049eb11be767", size = 277017, upload-time = "2026-06-28T19:55:06.29Z" }, + { url = "https://files.pythonhosted.org/packages/cb/e1/9eb83518e159d719fd681c4932dc2aaff855ce72451e1d05d69466f25a96/regex-2026.6.28-cp313-cp313t-macosx_10_13_universal2.whl", hash = "sha256:189dbf9fc4252d9f1352bf4bd1bef885edb6cc4b7341df202a65f821aaa3891c", size = 494195, upload-time = "2026-06-28T19:55:08.292Z" }, + { url = "https://files.pythonhosted.org/packages/fd/e2/e259c5f2f7be269d0e2fb54275c1fa6a13fb47019f389c3f3ae457447825/regex-2026.6.28-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:9277a4c6503390aa39cb4483b87ec0384faee0850a23b5cea33d008b5d8d83f1", size = 293976, upload-time = "2026-06-28T19:55:10.014Z" }, + { url = "https://files.pythonhosted.org/packages/8d/4e/9bdf444014d22b045d0c82ca114fac7e07a597b5b5331b7c4ce6328426e2/regex-2026.6.28-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:17eddca4e8ea9af0b5739314776cdf0172a49731ab61f2e1ea66e066ddd46c97", size = 292340, upload-time = "2026-06-28T19:55:11.88Z" }, + { url = "https://files.pythonhosted.org/packages/fd/3a/f49b11e59cbfe187ace0053a460bd72a0169b8cd52e7db9421a074ce7a43/regex-2026.6.28-cp313-cp313t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:e4466b8641e00c697aab5a73150150d2b2ea96b131c595691f42031abafd9f4d", size = 811704, upload-time = "2026-06-28T19:55:13.612Z" }, + { url = "https://files.pythonhosted.org/packages/2f/fb/ad04c39e149bf8b6cf357df5fff78341733ec366780a00c803a36735818c/regex-2026.6.28-cp313-cp313t-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:9cfcd4b0bdcf768c498415c170d1ed2a25a99bf0b65fa253bbd02f68ceba6475", size = 871157, upload-time = "2026-06-28T19:55:15.797Z" }, + { url = "https://files.pythonhosted.org/packages/7f/64/0e5ba31c11eb8ef7aac19a690c1211fc9aa9990caf09565785ebb0081b9a/regex-2026.6.28-cp313-cp313t-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:80c7adf1ef647f6b1e8aa2ca280e517174cd08bdf7a2e412cdfb68bd6a0917cb", size = 917287, upload-time = "2026-06-28T19:55:18.692Z" }, + { url = "https://files.pythonhosted.org/packages/11/75/6b78df2b858c2fcbbc4858fdc3f2975cf2703be374b2842db7d2c32591a7/regex-2026.6.28-cp313-cp313t-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:a043f5770e82283a22aed4cefef1a4e0f9dd8fd7184cb6ce0ad2e579e2134a9e", size = 816333, upload-time = "2026-06-28T19:55:20.973Z" }, + { url = "https://files.pythonhosted.org/packages/b4/01/ecfe665a3694d5eda9f3ec686c856438ada0943947b6005e90556a1e2cdf/regex-2026.6.28-cp313-cp313t-manylinux_2_31_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:3bd630a8dba06b55254ea5ee862194edab52ec783100d2ef1cd15a9c512fee27", size = 785518, upload-time = "2026-06-28T19:55:23.003Z" }, + { url = "https://files.pythonhosted.org/packages/b4/0a/88f9cd88ff1e82881605c4ffd62d77ee67d051232cfe6f8e9a64b86cf0e8/regex-2026.6.28-cp313-cp313t-musllinux_1_2_aarch64.whl", hash = "sha256:b77207e3cee13086f1906a6a2a12b41244c577e8ad9370d4b35ae1d548d354f3", size = 801371, upload-time = "2026-06-28T19:55:24.888Z" }, + { url = "https://files.pythonhosted.org/packages/a8/97/601483732f93275482ceb9fed57813dfed7c47d3a019db6ec4a3bb6e23e0/regex-2026.6.28-cp313-cp313t-musllinux_1_2_ppc64le.whl", hash = "sha256:6de82c268e5d101ee9e3ffd869924aa9a371e3a21e752cf4fa17b6ce50d219f7", size = 866517, upload-time = "2026-06-28T19:55:27.232Z" }, + { url = "https://files.pythonhosted.org/packages/81/ed/385c2a0351b994a693453c1d1a6e9af9eb35db3c9460d76b5078acd70c62/regex-2026.6.28-cp313-cp313t-musllinux_1_2_riscv64.whl", hash = "sha256:b15859e3908544fb99cf47341dcf0bfd089147d258c4c4d8a29e5b087f8085cb", size = 772834, upload-time = "2026-06-28T19:55:29.154Z" }, + { url = "https://files.pythonhosted.org/packages/06/bc/bbf4a5b3b29770d7f307d3c28b5b1bca0105b0cb424be0a4eb1339bc92cf/regex-2026.6.28-cp313-cp313t-musllinux_1_2_s390x.whl", hash = "sha256:c91487a917edd48a1ea646fdf60d7936d304f0e686fa7ea8326e47efca51d816", size = 856606, upload-time = "2026-06-28T19:55:32.186Z" }, + { url = "https://files.pythonhosted.org/packages/28/26/51d74fff82f682819979249f8d700267108ba5dc4eb284b0e11b9c85e4b3/regex-2026.6.28-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:c4ac65f3e3a99fd8f3a4a74e7a6610acd1ce9dfe9b8a03d346a4922380d68aeb", size = 803475, upload-time = "2026-06-28T19:55:34.328Z" }, + { url = "https://files.pythonhosted.org/packages/7c/3e/6be10cefdc813533fe604dbf5d3c77d2638e7ee658b2749ebadc113b6b2e/regex-2026.6.28-cp313-cp313t-win32.whl", hash = "sha256:3f6316f258bc7e6c9c2acbe9954947bbd397a81be3742a637a555f1855d6618d", size = 269126, upload-time = "2026-06-28T19:55:36.565Z" }, + { url = "https://files.pythonhosted.org/packages/3c/3c/32cda905ea1a6eeeb798291c294d8ec66ee0efe0cdba28b061e248b1d396/regex-2026.6.28-cp313-cp313t-win_amd64.whl", hash = "sha256:1484bdd6fba28422df9b5ebb04055b2e1b680e8e4f08490bb21ff0f3cc50d0ab", size = 279961, upload-time = "2026-06-28T19:55:38.456Z" }, + { url = "https://files.pythonhosted.org/packages/ea/b9/69f4e5cd6fbe0bb420cb2dbae441ca118f2495bdda522a74da75aa9829e7/regex-2026.6.28-cp313-cp313t-win_arm64.whl", hash = "sha256:3f15020f0b69cafe57baa067ff65b29acef68ff6b1670a53bef1ca11d708e02d", size = 279266, upload-time = "2026-06-28T19:55:40.62Z" }, +] + [[package]] name = "requests" version = "2.33.1" @@ -3716,6 +3864,30 @@ wheels = [ { url = "https://files.pythonhosted.org/packages/58/ed/dea90a65b7d9e69888890fb14c90d7f51bf0c1e82ad800aeb0160e4bacfd/ruff-0.15.10-py3-none-win_arm64.whl", hash = "sha256:601d1610a9e1f1c2165a4f561eeaa2e2ea1e97f3287c5aa258d3dab8b57c6188", size = 11035607, upload-time = "2026-04-09T14:05:47.593Z" }, ] +[[package]] +name = "safetensors" +version = "0.8.0" +source = { registry = "https://pypi.org/simple" } +sdist = { url = "https://files.pythonhosted.org/packages/45/06/f955dbbb1859e3bd23c8ac6141af5106e7ad5fedec4a3a6e3d60f94b7001/safetensors-0.8.0.tar.gz", hash = "sha256:fabaf3e0f18a6618d9b36560682562157f77c2b71fcffc7b432be2baed9d753d", size = 325846, upload-time = "2026-06-09T07:52:25.563Z" } +wheels = [ + { url = "https://files.pythonhosted.org/packages/39/a0/f718cda65b05407d228f97602cf60dca269c979867aa5beb25410de26cd3/safetensors-0.8.0-cp310-abi3-macosx_10_12_x86_64.whl", hash = "sha256:c554f85858e05226d3c2828e32395e677434685d6d94594a41643361c5e837f0", size = 473568, upload-time = "2026-06-09T07:52:18.829Z" }, + { url = "https://files.pythonhosted.org/packages/f5/b1/fa7c600e7dceae12e9606c7578cbc9ff1e1ed55844883ee5c92205e86226/safetensors-0.8.0-cp310-abi3-macosx_11_0_arm64.whl", hash = "sha256:c80201d22cbf405b80647a60ada77bba06c8fba2da2743ba1e89cdcc39a81f25", size = 484562, upload-time = "2026-06-09T07:52:17.518Z" }, + { url = "https://files.pythonhosted.org/packages/09/7d/65a7de0af421317bb36a067241e4235fff194eed60b961ed6d3f59a3fc60/safetensors-0.8.0-cp310-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:7a46e5ff292c356d6991e60942ba7f79817682d3a2cef0702136448cb9c4d235", size = 502844, upload-time = "2026-06-09T07:52:07.624Z" }, + { url = "https://files.pythonhosted.org/packages/91/4f/3175c9d75634e0e0dda0082794193521035edd7c70a6f212bf33ca06ddf4/safetensors-0.8.0-cp310-abi3-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:4124502b78f03534117c848f87a39b8f31e577b15eff423bf8bfb95f2a8c30d0", size = 511823, upload-time = "2026-06-09T07:52:09.565Z" }, + { url = "https://files.pythonhosted.org/packages/20/87/846c289e7aa2299eff406335717cf43ce8777194ece8aad75772e0411615/safetensors-0.8.0-cp310-abi3-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:7bc0a787ba8a35be368ee3574edfa2b1ad389eebd0a72e482ae275490e3f6c98", size = 633461, upload-time = "2026-06-09T07:52:11.128Z" }, + { url = "https://files.pythonhosted.org/packages/76/22/8d64d9df2c45d5ded401df889d0ad90882804ca172d79ec4f0df8f727fe0/safetensors-0.8.0-cp310-abi3-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:040070828e36dc8e122178bbbd5830ff9e97920affb84cbe0f46442497bed358", size = 545148, upload-time = "2026-06-09T07:52:13.603Z" }, + { url = "https://files.pythonhosted.org/packages/28/50/f203ff3a3ddfe19308efc83c5a3a29ed02bf786732ec35e68bf9162f3365/safetensors-0.8.0-cp310-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:fd6f3f93c9a0a7cc2788ee63fb763353d4bd2e89b0751bc78fcf7dda00bea774", size = 516040, upload-time = "2026-06-09T07:52:16.29Z" }, + { url = "https://files.pythonhosted.org/packages/46/fb/cdaed17ceb2948784fd9c36b6fd3e951b608547cea81a48e8ee6f8cfdfcb/safetensors-0.8.0-cp310-abi3-manylinux_2_31_riscv64.whl", hash = "sha256:fcdd41ec4628fee5799f807c73c353629130fbd942aa23d83c623dd6c9d52d78", size = 513832, upload-time = "2026-06-09T07:52:12.37Z" }, + { url = "https://files.pythonhosted.org/packages/0d/49/1e15de264dcc3b77943d2d0c56a95809956883b1c2d6d585c792523f180b/safetensors-0.8.0-cp310-abi3-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:8e9f537aa183a38ace122d27303dcd986b26bd2a7591f9181d7f0c396f4677ca", size = 559930, upload-time = "2026-06-09T07:52:14.743Z" }, + { url = "https://files.pythonhosted.org/packages/2a/43/bf38443278eab4b1be1fce2931e2b012ad9cb7df52ada751d0aab8f7659a/safetensors-0.8.0-cp310-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:87eec7ffed2b809f05a398a8becb7d013f19f7837cd15d9748580d6cf30dbaf4", size = 678670, upload-time = "2026-06-09T07:52:20.032Z" }, + { url = "https://files.pythonhosted.org/packages/72/e3/68cd3fa5b48488e84add63e04cb12f3bc28ae4638c06d4508c6e88823d0e/safetensors-0.8.0-cp310-abi3-musllinux_1_2_armv7l.whl", hash = "sha256:4a95ae2b05d7726d751da4ebf626a2ca782b706e101bd894c95bc2450b1cffcc", size = 786679, upload-time = "2026-06-09T07:52:21.322Z" }, + { url = "https://files.pythonhosted.org/packages/29/4b/1c19c509d56e01f4fbb3d0a2e597450f6cc04d1d56cf52defb0a62dfd715/safetensors-0.8.0-cp310-abi3-musllinux_1_2_i686.whl", hash = "sha256:3ae091f16662658bdc019a4ff6cb4c085bb7d725eb5978b183ffd265863b6d2d", size = 765683, upload-time = "2026-06-09T07:52:22.594Z" }, + { url = "https://files.pythonhosted.org/packages/27/43/41c1621732edd934d868a00d1b891584c892a7b62a9aab82ea5a0a5623ee/safetensors-0.8.0-cp310-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:8e080062fcde23be189565e1c3305d16751a218ecf9412c8601e64204eb6f846", size = 722361, upload-time = "2026-06-09T07:52:23.924Z" }, + { url = "https://files.pythonhosted.org/packages/8e/3f/73ccf82579412b4a71c4ca673f10b5f1f888d7cf5af7fe24f27d30307be4/safetensors-0.8.0-cp310-abi3-win32.whl", hash = "sha256:2ddf52eac562eda224f99acfa7889d02968c1fd59a5b011ae7d8137c37e9c02d", size = 342401, upload-time = "2026-06-09T07:52:28.895Z" }, + { url = "https://files.pythonhosted.org/packages/1b/6d/3fba214c1e5e0f69991677ec3bc17023f0421776975e1de0c682dca475e2/safetensors-0.8.0-cp310-abi3-win_amd64.whl", hash = "sha256:096ec1a98435df7beb08853bb5aa9081a84f23d0adc67ed1a0a10550f608373f", size = 355540, upload-time = "2026-06-09T07:52:27.832Z" }, + { url = "https://files.pythonhosted.org/packages/8d/fc/7eedc3510d97878876e32774eebbeb61c43f148a96e915c84229a3e967aa/safetensors-0.8.0-cp310-abi3-win_arm64.whl", hash = "sha256:f7838e5135a406ad3e02efdcb8cf2e5397d368b0154537c4fec682dbc544d452", size = 340500, upload-time = "2026-06-09T07:52:26.745Z" }, +] + [[package]] name = "scikit-image" version = "0.26.0" @@ -4036,6 +4208,15 @@ wheels = [ { url = "https://files.pythonhosted.org/packages/79/d5/1ec3120f461f31a03d1d2f1d339f5058f12c7a542d22bfcc350511eccc8a/shap-0.51.0-cp313-cp313-win_amd64.whl", hash = "sha256:5f51ca55bda10b3fa2125f2b8e08e9d6a6edcbb0e752c67050e87a6d3ca7d53b", size = 555927, upload-time = "2026-03-04T09:18:01.274Z" }, ] +[[package]] +name = "shellingham" +version = "1.5.4" +source = { registry = "https://pypi.org/simple" } +sdist = { url = "https://files.pythonhosted.org/packages/58/15/8b3609fd3830ef7b27b655beb4b4e9c62313a4e8da8c676e142cc210d58e/shellingham-1.5.4.tar.gz", hash = "sha256:8dbca0739d487e5bd35ab3ca4b36e11c4078f3a234bfce294b0a0291363404de", size = 10310, upload-time = "2023-10-24T04:13:40.426Z" } +wheels = [ + { url = "https://files.pythonhosted.org/packages/e0/f9/0595336914c5619e5f28a1fb793285925a8cd4b432c9da0a987836c7f822/shellingham-1.5.4-py2.py3-none-any.whl", hash = "sha256:7ecfff8f2fd72616f7481040475a65b2bf8af90a56c89140852d1120324e8686", size = 9755, upload-time = "2023-10-24T04:13:38.866Z" }, +] + [[package]] name = "six" version = "1.17.0" @@ -4547,6 +4728,32 @@ wheels = [ { url = "https://files.pythonhosted.org/packages/1f/6e/7d8850ff112f8f80d394ca45e89b975a3a43559d47af3137b767669b3294/tifffile-2026.5.15-py3-none-any.whl", hash = "sha256:6715515a53cabc0cefc5c9f13a0ae2c250e63e2ca784ce02d0b6c333810c2a17", size = 266665, upload-time = "2026-05-15T20:04:54.227Z" }, ] +[[package]] +name = "tokenizers" +version = "0.22.2" +source = { registry = "https://pypi.org/simple" } +dependencies = [ + { name = "huggingface-hub" }, +] +sdist = { url = "https://files.pythonhosted.org/packages/73/6f/f80cfef4a312e1fb34baf7d85c72d4411afde10978d4657f8cdd811d3ccc/tokenizers-0.22.2.tar.gz", hash = "sha256:473b83b915e547aa366d1eee11806deaf419e17be16310ac0a14077f1e28f917", size = 372115, upload-time = "2026-01-05T10:45:15.988Z" } +wheels = [ + { url = "https://files.pythonhosted.org/packages/92/97/5dbfabf04c7e348e655e907ed27913e03db0923abb5dfdd120d7b25630e1/tokenizers-0.22.2-cp39-abi3-macosx_10_12_x86_64.whl", hash = "sha256:544dd704ae7238755d790de45ba8da072e9af3eea688f698b137915ae959281c", size = 3100275, upload-time = "2026-01-05T10:41:02.158Z" }, + { url = "https://files.pythonhosted.org/packages/2e/47/174dca0502ef88b28f1c9e06b73ce33500eedfac7a7692108aec220464e7/tokenizers-0.22.2-cp39-abi3-macosx_11_0_arm64.whl", hash = "sha256:1e418a55456beedca4621dbab65a318981467a2b188e982a23e117f115ce5001", size = 2981472, upload-time = "2026-01-05T10:41:00.276Z" }, + { url = "https://files.pythonhosted.org/packages/d6/84/7990e799f1309a8b87af6b948f31edaa12a3ed22d11b352eaf4f4b2e5753/tokenizers-0.22.2-cp39-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:2249487018adec45d6e3554c71d46eb39fa8ea67156c640f7513eb26f318cec7", size = 3290736, upload-time = "2026-01-05T10:40:32.165Z" }, + { url = "https://files.pythonhosted.org/packages/78/59/09d0d9ba94dcd5f4f1368d4858d24546b4bdc0231c2354aa31d6199f0399/tokenizers-0.22.2-cp39-abi3-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:25b85325d0815e86e0bac263506dd114578953b7b53d7de09a6485e4a160a7dd", size = 3168835, upload-time = "2026-01-05T10:40:38.847Z" }, + { url = "https://files.pythonhosted.org/packages/47/50/b3ebb4243e7160bda8d34b731e54dd8ab8b133e50775872e7a434e524c28/tokenizers-0.22.2-cp39-abi3-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:bfb88f22a209ff7b40a576d5324bf8286b519d7358663db21d6246fb17eea2d5", size = 3521673, upload-time = "2026-01-05T10:40:56.614Z" }, + { url = "https://files.pythonhosted.org/packages/e0/fa/89f4cb9e08df770b57adb96f8cbb7e22695a4cb6c2bd5f0c4f0ebcf33b66/tokenizers-0.22.2-cp39-abi3-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:1c774b1276f71e1ef716e5486f21e76333464f47bece56bbd554485982a9e03e", size = 3724818, upload-time = "2026-01-05T10:40:44.507Z" }, + { url = "https://files.pythonhosted.org/packages/64/04/ca2363f0bfbe3b3d36e95bf67e56a4c88c8e3362b658e616d1ac185d47f2/tokenizers-0.22.2-cp39-abi3-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:df6c4265b289083bf710dff49bc51ef252f9d5be33a45ee2bed151114a56207b", size = 3379195, upload-time = "2026-01-05T10:40:51.139Z" }, + { url = "https://files.pythonhosted.org/packages/2e/76/932be4b50ef6ccedf9d3c6639b056a967a86258c6d9200643f01269211ca/tokenizers-0.22.2-cp39-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:369cc9fc8cc10cb24143873a0d95438bb8ee257bb80c71989e3ee290e8d72c67", size = 3274982, upload-time = "2026-01-05T10:40:58.331Z" }, + { url = "https://files.pythonhosted.org/packages/1d/28/5f9f5a4cc211b69e89420980e483831bcc29dade307955cc9dc858a40f01/tokenizers-0.22.2-cp39-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:29c30b83d8dcd061078b05ae0cb94d3c710555fbb44861139f9f83dcca3dc3e4", size = 9478245, upload-time = "2026-01-05T10:41:04.053Z" }, + { url = "https://files.pythonhosted.org/packages/6c/fb/66e2da4704d6aadebf8cb39f1d6d1957df667ab24cff2326b77cda0dcb85/tokenizers-0.22.2-cp39-abi3-musllinux_1_2_armv7l.whl", hash = "sha256:37ae80a28c1d3265bb1f22464c856bd23c02a05bb211e56d0c5301a435be6c1a", size = 9560069, upload-time = "2026-01-05T10:45:10.673Z" }, + { url = "https://files.pythonhosted.org/packages/16/04/fed398b05caa87ce9b1a1bb5166645e38196081b225059a6edaff6440fac/tokenizers-0.22.2-cp39-abi3-musllinux_1_2_i686.whl", hash = "sha256:791135ee325f2336f498590eb2f11dc5c295232f288e75c99a36c5dbce63088a", size = 9899263, upload-time = "2026-01-05T10:45:12.559Z" }, + { url = "https://files.pythonhosted.org/packages/05/a1/d62dfe7376beaaf1394917e0f8e93ee5f67fea8fcf4107501db35996586b/tokenizers-0.22.2-cp39-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:38337540fbbddff8e999d59970f3c6f35a82de10053206a7562f1ea02d046fa5", size = 10033429, upload-time = "2026-01-05T10:45:14.333Z" }, + { url = "https://files.pythonhosted.org/packages/fd/18/a545c4ea42af3df6effd7d13d250ba77a0a86fb20393143bbb9a92e434d4/tokenizers-0.22.2-cp39-abi3-win32.whl", hash = "sha256:a6bf3f88c554a2b653af81f3204491c818ae2ac6fbc09e76ef4773351292bc92", size = 2502363, upload-time = "2026-01-05T10:45:20.593Z" }, + { url = "https://files.pythonhosted.org/packages/65/71/0670843133a43d43070abeb1949abfdef12a86d490bea9cd9e18e37c5ff7/tokenizers-0.22.2-cp39-abi3-win_amd64.whl", hash = "sha256:c9ea31edff2968b44a88f97d784c2f16dc0729b8b143ed004699ebca91f05c48", size = 2747786, upload-time = "2026-01-05T10:45:18.411Z" }, + { url = "https://files.pythonhosted.org/packages/72/f4/0de46cfa12cdcbcd464cc59fde36912af405696f687e53a091fb432f694c/tokenizers-0.22.2-cp39-abi3-win_arm64.whl", hash = "sha256:9ce725d22864a1e965217204946f830c37876eee3b2ba6fc6255e8e903d5fcbc", size = 2612133, upload-time = "2026-01-05T10:45:17.232Z" }, +] + [[package]] name = "tomli" version = "2.4.1" @@ -5057,6 +5264,26 @@ wheels = [ { url = "https://files.pythonhosted.org/packages/16/e1/3079a9ff9b8e11b846c6ac5c8b5bfb7ff225eee721825310c91b3b50304f/tqdm-4.67.3-py3-none-any.whl", hash = "sha256:ee1e4c0e59148062281c49d80b25b67771a127c85fc9676d3be5f243206826bf", size = 78374, upload-time = "2026-02-03T17:35:50.982Z" }, ] +[[package]] +name = "transformers" +version = "5.13.0" +source = { registry = "https://pypi.org/simple" } +dependencies = [ + { name = "huggingface-hub" }, + { name = "numpy" }, + { name = "packaging" }, + { name = "pyyaml" }, + { name = "regex" }, + { name = "safetensors" }, + { name = "tokenizers" }, + { name = "tqdm" }, + { name = "typer" }, +] +sdist = { url = "https://files.pythonhosted.org/packages/ec/e1/720ff7ff666b04279fea5bb7ac3ef8675e98f0ddbc1b8cb8bc9f3889d62e/transformers-5.13.0.tar.gz", hash = "sha256:940c1428e42a4238f9ccf0cd41e63c590701aa63c19fd2ce3d7d602222d68495", size = 9195801, upload-time = "2026-07-03T16:05:39.362Z" } +wheels = [ + { url = "https://files.pythonhosted.org/packages/a5/3a/d99704c5effe10c6339c98cb236259161103e159bb99a78468b6729572ec/transformers-5.13.0-py3-none-any.whl", hash = "sha256:8adbc1d20bd5463cd6876b2eb7cb31971e1065788e7dc6bc12bab597a7c504b7", size = 11503730, upload-time = "2026-07-03T16:05:35.569Z" }, +] + [[package]] name = "triton" version = "3.4.0" @@ -5089,6 +5316,21 @@ wheels = [ { url = "https://download-r2.pytorch.org/whl/triton_xpu-3.7.0-cp313-cp313t-win_amd64.whl", hash = "sha256:de55ca5cbb2b356a1640e3ec5af0c42c98b1d3ee925c6a6af590975fe09cc156", upload-time = "2026-05-12T14:53:01Z" }, ] +[[package]] +name = "typer" +version = "0.26.8" +source = { registry = "https://pypi.org/simple" } +dependencies = [ + { name = "annotated-doc" }, + { name = "colorama", marker = "sys_platform == 'win32' or (extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-onnx-intel') or (extra == 'extra-6-raitap-auto-lirpa' and extra == 'extra-6-raitap-torch-intel') or (extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-cuda') or (extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-onnx-intel') or (extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-cuda') or (extra == 'extra-6-raitap-onnx-cpu' and extra == 'extra-6-raitap-torch-intel') or (extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-onnx-intel') or (extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-cpu') or (extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-torch-intel') or (extra == 'extra-6-raitap-onnx-cuda' and extra == 'extra-6-raitap-xgboost') or (extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cpu') or (extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-torch-cuda') or (extra == 'extra-6-raitap-onnx-intel' and extra == 'extra-6-raitap-xgboost') or (extra == 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-cuda') or (extra == 'extra-6-raitap-torch-cpu' and extra == 'extra-6-raitap-torch-intel') or (extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-torch-intel') or (extra == 'extra-6-raitap-torch-cuda' and extra == 'extra-6-raitap-xgboost') or (extra == 'extra-6-raitap-torch-intel' and extra == 'extra-6-raitap-xgboost')" }, + { name = "rich" }, + { name = "shellingham" }, +] +sdist = { url = "https://files.pythonhosted.org/packages/7c/f7/68adc395201b20b872d68e975386832e8005ffeacedd43a1d837a32815be/typer-0.26.8.tar.gz", hash = "sha256:c244a6bd558886fe3f8780efb6bdd28bb9aff005a94eedebaa5cb32926fe2f7e", size = 202097, upload-time = "2026-06-26T09:22:45.705Z" } +wheels = [ + { url = "https://files.pythonhosted.org/packages/80/87/b9fd69c92c6102a066e1b86a35243f53e70bd4c709f2a26d9f4fee4f4dc0/typer-0.26.8-py3-none-any.whl", hash = "sha256:3512ca79ac5c11113414b36e80281b872884477722440691c89d1112e321a49c", size = 122564, upload-time = "2026-06-26T09:22:44.72Z" }, +] + [[package]] name = "typing-extensions" version = "4.15.0" From 7f06da70781594949ced16f0d3c8648440a63f08 Mon Sep 17 00:00:00 2001 From: Stanislas Laurent Date: Sun, 5 Jul 2026 20:56:13 +0200 Subject: [PATCH 03/24] feat(data): add pluggable input-parser registry + shared parser factory (refs #340) --- src/raitap/configs/schema.py | 6 ++ src/raitap/configs/zen.py | 1 + src/raitap/data/_parser_factory.py | 25 +++++++ src/raitap/data/input_parsers/__init__.py | 5 ++ src/raitap/data/input_parsers/base.py | 16 +++++ src/raitap/data/input_parsers/factory.py | 12 ++++ src/raitap/data/input_parsers/registration.py | 27 +++++++ src/raitap/data/label_parsers/factory.py | 30 ++------ .../data/tests/test_input_parser_registry.py | 71 +++++++++++++++++++ 9 files changed, 167 insertions(+), 26 deletions(-) create mode 100644 src/raitap/data/_parser_factory.py create mode 100644 src/raitap/data/input_parsers/__init__.py create mode 100644 src/raitap/data/input_parsers/base.py create mode 100644 src/raitap/data/input_parsers/factory.py create mode 100644 src/raitap/data/input_parsers/registration.py create mode 100644 src/raitap/data/tests/test_input_parser_registry.py diff --git a/src/raitap/configs/schema.py b/src/raitap/configs/schema.py index 8820e5b5..e5473a15 100644 --- a/src/raitap/configs/schema.py +++ b/src/raitap/configs/schema.py @@ -75,6 +75,11 @@ class LabelsConfig: _target_: str = MISSING +@dataclass +class InputsConfig: + _target_: str = MISSING + + @dataclass class TabularLabelsConfig(LabelsConfig): _target_: str = "TabularLabelParser" @@ -153,6 +158,7 @@ class DataConfig: # modality for non-image data such as ACAS Xu's 5-feature tabular vector. input_metadata: dict[str, Any] | None = None labels: LabelsConfig | None = None + inputs: InputsConfig | None = None @dataclass diff --git a/src/raitap/configs/zen.py b/src/raitap/configs/zen.py index c0ac3928..05a893cf 100644 --- a/src/raitap/configs/zen.py +++ b/src/raitap/configs/zen.py @@ -49,6 +49,7 @@ def register_zen_groups() -> None: import importlib for pkg in ( + "raitap.data.input_parsers", "raitap.data.label_parsers", "raitap.metrics", "raitap.reporting", diff --git a/src/raitap/data/_parser_factory.py b/src/raitap/data/_parser_factory.py new file mode 100644 index 00000000..4d9a552b --- /dev/null +++ b/src/raitap/data/_parser_factory.py @@ -0,0 +1,25 @@ +"""Shared _target_-resolve-and-instantiate core for data parser families.""" + +from __future__ import annotations + +from typing import Any + +from hydra.utils import instantiate + +from raitap import raitap_log +from raitap.configs import cfg_to_dict, resolve_target + + +def create_parser(config: Any, *, prefix: str, kind: str) -> Any: + """Instantiate a parser from Hydra-style config (``_target_`` + kwargs).""" + cfg = cfg_to_dict(config) + target_path: str = cfg.get("_target_", "") + cfg["_target_"] = resolve_target(target_path, prefix) + try: + return instantiate(cfg) + except Exception as e: + raitap_log.exception("%s instantiation failed for target %r", kind, target_path) + raise ValueError( + f"Could not instantiate {kind} {target_path!r}.\n" + f"Check that _target_ points to a valid {kind} implementation." + ) from e diff --git a/src/raitap/data/input_parsers/__init__.py b/src/raitap/data/input_parsers/__init__.py new file mode 100644 index 00000000..f02a24be --- /dev/null +++ b/src/raitap/data/input_parsers/__init__.py @@ -0,0 +1,5 @@ +"""Input-parser family package. Importing fires @input_parser decorators.""" + +from __future__ import annotations + +__all__: list[str] = [] diff --git a/src/raitap/data/input_parsers/base.py b/src/raitap/data/input_parsers/base.py new file mode 100644 index 00000000..7d8d749d --- /dev/null +++ b/src/raitap/data/input_parsers/base.py @@ -0,0 +1,16 @@ +"""Base protocol for input-source parsers (bytes/column -> list[str]).""" + +from __future__ import annotations + +from typing import Protocol, runtime_checkable + +from raitap.data.types import InputModality # noqa: TC001 runtime-resolvable for get_type_hints() + + +@runtime_checkable +class InputParser(Protocol): + """Protocol every input-source parser must satisfy.""" + + supported_modalities: frozenset[InputModality] + + def parse(self, *, source: str, sample_ids: list[str] | None) -> list[str]: ... diff --git a/src/raitap/data/input_parsers/factory.py b/src/raitap/data/input_parsers/factory.py new file mode 100644 index 00000000..0ff14c9c --- /dev/null +++ b/src/raitap/data/input_parsers/factory.py @@ -0,0 +1,12 @@ +"""Instantiation factory for input parsers (delegates to _parser_factory).""" + +from __future__ import annotations + +from typing import Any + +from raitap.data._parser_factory import create_parser +from raitap.data.input_parsers.base import InputParser # noqa: TC001 runtime-resolvable + + +def create_input_parser(inputs_config: Any) -> InputParser: + return create_parser(inputs_config, prefix="raitap.data.input_parsers.", kind="input parser") diff --git a/src/raitap/data/input_parsers/registration.py b/src/raitap/data/input_parsers/registration.py new file mode 100644 index 00000000..a83c759c --- /dev/null +++ b/src/raitap/data/input_parsers/registration.py @@ -0,0 +1,27 @@ +"""Family decorator for input-parser adapters. Mirrors label_parsers/registration.py.""" + +from __future__ import annotations + +from typing import TYPE_CHECKING, TypeVar, Unpack + +from raitap._adapters import AdapterDecoratorOptions, FamilyConfig, _register_core +from raitap.configs.schema import InputsConfig + +if TYPE_CHECKING: + from collections.abc import Callable + + from raitap.data.input_parsers.base import InputParser + +INPUTS = FamilyConfig(group="data/inputs", schema=InputsConfig, package_style="flat") + +T = TypeVar("T", bound="InputParser") + + +def input_parser(**common: Unpack[AdapterDecoratorOptions]) -> Callable[[type[T]], type[T]]: + """Register an input-parser adapter. ``registry_name`` required.""" + common.setdefault("extra", "") + + def wrap(cls: type[T]) -> type[T]: + return _register_core(cls, family=INPUTS, **common) + + return wrap diff --git a/src/raitap/data/label_parsers/factory.py b/src/raitap/data/label_parsers/factory.py index 0ec4b8e7..9758055d 100644 --- a/src/raitap/data/label_parsers/factory.py +++ b/src/raitap/data/label_parsers/factory.py @@ -1,34 +1,12 @@ -"""Instantiation factory for label parsers (mirrors metrics/factory.py:44-60).""" +"""Instantiation factory for label parsers (delegates to _parser_factory).""" from __future__ import annotations from typing import Any -from hydra.utils import instantiate - -from raitap import raitap_log -from raitap.configs import cfg_to_dict, resolve_target -from raitap.data.label_parsers.base import ( - LabelParser, # noqa: TC001 must stay runtime-resolvable for get_type_hints() -) - -_LABELS_PREFIX = "raitap.data.label_parsers." +from raitap.data._parser_factory import create_parser +from raitap.data.label_parsers.base import LabelParser # noqa: TC001 runtime-resolvable def create_label_parser(labels_config: Any) -> LabelParser: - """Instantiate a label parser from Hydra-style config (``_target_`` + kwargs).""" - labels_cfg = cfg_to_dict(labels_config) - target_path: str = labels_cfg.get("_target_", "") - resolved_target = resolve_target(target_path, _LABELS_PREFIX) - labels_cfg["_target_"] = resolved_target - - try: - parser = instantiate(labels_cfg) - except Exception as e: - raitap_log.exception("Label parser instantiation failed for target %r", target_path) - raise ValueError( - f"Could not instantiate label parser {target_path!r}.\n" - "Check that _target_ points to a valid LabelParser implementation." - ) from e - - return parser + return create_parser(labels_config, prefix="raitap.data.label_parsers.", kind="label parser") diff --git a/src/raitap/data/tests/test_input_parser_registry.py b/src/raitap/data/tests/test_input_parser_registry.py new file mode 100644 index 00000000..deeb7406 --- /dev/null +++ b/src/raitap/data/tests/test_input_parser_registry.py @@ -0,0 +1,71 @@ +from raitap.configs.schema import InputsConfig +from raitap.data.input_parsers.base import InputParser +from raitap.data.input_parsers.registration import input_parser +from raitap.data.label_parsers.factory import create_label_parser +from raitap.data.label_parsers.tabular import TabularLabelParser +from raitap.data.types import InputModality + +# Module-level (importable qualname) so hydra-zen's ``builds()`` can resolve a +# ``_target_`` for it; a class defined inline inside a test function has no +# importable qualname and ``_register_core`` silently skips ConfigStore +# registration for it (see ``raitap._adapters._register_core``). +_COMPOSED_TARGET = "raitap.data.tests.test_input_parser_registry.DummyComposeInputParser" + + +@input_parser(registry_name="dummy_compose", schema=InputsConfig) +class DummyComposeInputParser: + supported_modalities = frozenset({InputModality.text}) + + def parse(self, *, source: str, sample_ids: list[str] | None) -> list[str]: + return ["a", "b"] + + +def _register_inputs_group() -> None: + """Register the ``data/inputs`` group via the canonical ``register_configs``. + + Mirrors ``configs/tests/test_labels_schema.py::_register_labels_group``: + uses the same registration path as production (and the rest of the suite) + rather than a raw ``store.add_to_hydra_store(overwrite_ok=True)``, which + would clobber other groups' schema nodes. + """ + from raitap.configs import register_configs + + register_configs() + + +def test_integration_compose_data_inputs_dummy() -> None: + """Composing +data/inputs=dummy_compose lands cfg.data.inputs at the FQN. + + Regression for the Critical review finding: ``DataConfig`` was missing an + ``inputs`` field, so composing any ``data/inputs`` variant raised + ``ConfigKeyError: Key 'inputs' not in 'DataConfig'``. + """ + from hydra import compose, initialize + from hydra.core.global_hydra import GlobalHydra + + _register_inputs_group() + GlobalHydra.instance().clear() + with initialize(version_base=None, config_path=None): + cfg = compose(config_name="raitap_schema", overrides=["+data/inputs=dummy_compose"]) + assert cfg.data.inputs._target_ == _COMPOSED_TARGET + + +def test_decorated_parser_satisfies_protocol_and_registers() -> None: + @input_parser(registry_name="dummy_text", schema=InputsConfig) + class DummyParser: + supported_modalities = frozenset({InputModality.text}) + + def parse(self, *, source: str, sample_ids: list[str] | None) -> list[str]: + return ["a", "b"] + + p = DummyParser() + assert isinstance(p, InputParser) + assert InputModality.text in p.supported_modalities + assert p.parse(source="x", sample_ids=None) == ["a", "b"] + + +def test_create_label_parser_still_works_after_factory_refactor() -> None: + """Regression: the shared-core refactor of label_parsers/factory.py must + preserve existing label-parser instantiation behaviour.""" + parser = create_label_parser({"_target_": "TabularLabelParser", "source": "x"}) + assert isinstance(parser, TabularLabelParser) From 61924789cc2e5cb5eb0cae958832cd4f1a484276 Mon Sep 17 00:00:00 2001 From: Stanislas Laurent Date: Sun, 5 Jul 2026 21:02:12 +0200 Subject: [PATCH 04/24] feat(data): add text_csv/text_jsonl/text_dir input parsers (refs #340) --- src/raitap/configs/schema.py | 21 ++++++++++++++ src/raitap/data/input_parsers/__init__.py | 6 +++- src/raitap/data/input_parsers/text_csv.py | 23 +++++++++++++++ src/raitap/data/input_parsers/text_dir.py | 23 +++++++++++++++ src/raitap/data/input_parsers/text_jsonl.py | 25 +++++++++++++++++ .../data/tests/test_text_input_parsers.py | 28 +++++++++++++++++++ 6 files changed, 125 insertions(+), 1 deletion(-) create mode 100644 src/raitap/data/input_parsers/text_csv.py create mode 100644 src/raitap/data/input_parsers/text_dir.py create mode 100644 src/raitap/data/input_parsers/text_jsonl.py create mode 100644 src/raitap/data/tests/test_text_input_parsers.py diff --git a/src/raitap/configs/schema.py b/src/raitap/configs/schema.py index e5473a15..c962575f 100644 --- a/src/raitap/configs/schema.py +++ b/src/raitap/configs/schema.py @@ -80,6 +80,27 @@ class InputsConfig: _target_: str = MISSING +@dataclass +class TextCsvInputsConfig(InputsConfig): + _target_: str = "TextCsvInputParser" + source: str = MISSING + text_column: str = MISSING + id_column: str | None = None + + +@dataclass +class TextJsonlInputsConfig(InputsConfig): + _target_: str = "TextJsonlInputParser" + source: str = MISSING + text_field: str = "text" + + +@dataclass +class TextDirInputsConfig(InputsConfig): + _target_: str = "TextDirInputParser" + source: str = MISSING + + @dataclass class TabularLabelsConfig(LabelsConfig): _target_: str = "TabularLabelParser" diff --git a/src/raitap/data/input_parsers/__init__.py b/src/raitap/data/input_parsers/__init__.py index f02a24be..279a0426 100644 --- a/src/raitap/data/input_parsers/__init__.py +++ b/src/raitap/data/input_parsers/__init__.py @@ -2,4 +2,8 @@ from __future__ import annotations -__all__: list[str] = [] +from .text_csv import TextCsvInputParser # pyright: ignore[reportUnusedImport] +from .text_dir import TextDirInputParser # pyright: ignore[reportUnusedImport] +from .text_jsonl import TextJsonlInputParser # pyright: ignore[reportUnusedImport] + +__all__ = ["TextCsvInputParser", "TextDirInputParser", "TextJsonlInputParser"] diff --git a/src/raitap/data/input_parsers/text_csv.py b/src/raitap/data/input_parsers/text_csv.py new file mode 100644 index 00000000..7babc6ba --- /dev/null +++ b/src/raitap/data/input_parsers/text_csv.py @@ -0,0 +1,23 @@ +"""CSV/TSV/Parquet text input parser (reads one text column).""" + +from __future__ import annotations + +from raitap.configs.schema import TextCsvInputsConfig +from raitap.data.data import SourceKind, _load_tabular_frame, get_source_path +from raitap.data.input_parsers.registration import input_parser +from raitap.data.types import InputModality + + +@input_parser(registry_name="text_csv", schema=TextCsvInputsConfig) +class TextCsvInputParser: + supported_modalities = frozenset({InputModality.text}) + + def __init__(self, *, source: str, text_column: str, id_column: str | None = None) -> None: + self.source = source + self.text_column = text_column + self.id_column = id_column + + def parse(self, *, source: str, sample_ids: list[str] | None) -> list[str]: + path = get_source_path(self.source, kind=SourceKind.DATA) + frame = _load_tabular_frame(path) + return [str(v) for v in frame[self.text_column].tolist()] diff --git a/src/raitap/data/input_parsers/text_dir.py b/src/raitap/data/input_parsers/text_dir.py new file mode 100644 index 00000000..1f99c607 --- /dev/null +++ b/src/raitap/data/input_parsers/text_dir.py @@ -0,0 +1,23 @@ +"""Directory-of-.txt text input parser (one document per file, sorted by name).""" + +from __future__ import annotations + +from pathlib import Path + +from raitap.configs.schema import TextDirInputsConfig +from raitap.data.data import SourceKind, get_source_path +from raitap.data.input_parsers.registration import input_parser +from raitap.data.types import InputModality + + +@input_parser(registry_name="text_dir", schema=TextDirInputsConfig) +class TextDirInputParser: + supported_modalities = frozenset({InputModality.text}) + + def __init__(self, *, source: str) -> None: + self.source = source + + def parse(self, *, source: str, sample_ids: list[str] | None) -> list[str]: + root = Path(get_source_path(self.source, kind=SourceKind.DATA)) + files = sorted(root.glob("*.txt")) + return [f.read_text(encoding="utf-8") for f in files] diff --git a/src/raitap/data/input_parsers/text_jsonl.py b/src/raitap/data/input_parsers/text_jsonl.py new file mode 100644 index 00000000..1c5af22e --- /dev/null +++ b/src/raitap/data/input_parsers/text_jsonl.py @@ -0,0 +1,25 @@ +"""JSONL text input parser (one object per line, reads a text field).""" + +from __future__ import annotations + +import json +from pathlib import Path + +from raitap.configs.schema import TextJsonlInputsConfig +from raitap.data.data import SourceKind, get_source_path +from raitap.data.input_parsers.registration import input_parser +from raitap.data.types import InputModality + + +@input_parser(registry_name="text_jsonl", schema=TextJsonlInputsConfig) +class TextJsonlInputParser: + supported_modalities = frozenset({InputModality.text}) + + def __init__(self, *, source: str, text_field: str = "text") -> None: + self.source = source + self.text_field = text_field + + def parse(self, *, source: str, sample_ids: list[str] | None) -> list[str]: + path = get_source_path(self.source, kind=SourceKind.DATA) + lines = Path(path).read_text(encoding="utf-8").splitlines() + return [str(json.loads(line)[self.text_field]) for line in lines if line.strip()] diff --git a/src/raitap/data/tests/test_text_input_parsers.py b/src/raitap/data/tests/test_text_input_parsers.py new file mode 100644 index 00000000..b4fe2489 --- /dev/null +++ b/src/raitap/data/tests/test_text_input_parsers.py @@ -0,0 +1,28 @@ +import json +from pathlib import Path + +from raitap.data.input_parsers.text_csv import TextCsvInputParser +from raitap.data.input_parsers.text_dir import TextDirInputParser +from raitap.data.input_parsers.text_jsonl import TextJsonlInputParser + + +def test_csv_reads_text_column(tmp_path: Path) -> None: + p = tmp_path / "d.csv" + p.write_text("id,text\n0,hello\n1,world\n") + assert TextCsvInputParser(source=str(p), text_column="text").parse( + source=str(p), sample_ids=None + ) == ["hello", "world"] + + +def test_jsonl_reads_text_field(tmp_path: Path) -> None: + p = tmp_path / "d.jsonl" + p.write_text("\n".join(json.dumps({"text": t}) for t in ["a", "b"])) + assert TextJsonlInputParser(source=str(p)).parse(source=str(p), sample_ids=None) == ["a", "b"] + + +def test_dir_reads_txt_files_sorted(tmp_path: Path) -> None: + (tmp_path / "1.txt").write_text("one") + (tmp_path / "2.txt").write_text("two") + assert TextDirInputParser(source=str(tmp_path)).parse( + source=str(tmp_path), sample_ids=None + ) == ["one", "two"] From b3566a745c93141a69f0bc81bccc95c433bea819 Mon Sep 17 00:00:00 2001 From: Stanislas Laurent Date: Sun, 5 Jul 2026 21:12:36 +0200 Subject: [PATCH 05/24] feat(data): load + tokenise text into input_ids + attention_mask (refs #340) --- src/raitap/data/data.py | 66 +++++++++++++++++++++---- src/raitap/data/tests/test_load_text.py | 18 +++++++ 2 files changed, 74 insertions(+), 10 deletions(-) create mode 100644 src/raitap/data/tests/test_load_text.py diff --git a/src/raitap/data/data.py b/src/raitap/data/data.py index 8de717bb..b654c3c8 100644 --- a/src/raitap/data/data.py +++ b/src/raitap/data/data.py @@ -66,7 +66,8 @@ def __init__( self.tensor: torch.Tensor | DetectionInputs family = resolve_task_family(task_kind) self.input_modality: InputModality - raw_tensor, self.sample_ids, self.input_modality = self._load_data( + self.attention_mask: torch.Tensor | None = None + raw_tensor, self.sample_ids, self.input_modality, self.attention_mask = self._load_data( cfg, resolved_preprocessing=resolved_preprocessing, ) @@ -83,7 +84,9 @@ def _load_data( cfg: AppConfig, *, resolved_preprocessing: ResolvedPreprocessing | None = None, - ) -> tuple[torch.Tensor | DetectionInputs, list[str] | None, InputModality]: + ) -> tuple[ + torch.Tensor | DetectionInputs, list[str] | None, InputModality, torch.Tensor | None + ]: """ Load data from a specified source into a raw tensor. @@ -97,16 +100,20 @@ def _load_data( - Local image file → ``(1, 3, H, W)`` float32 in ``[0, 1]`` (or ``list[Tensor]`` of length 1 for detection) - Local CSV / TSV / Parquet file → ``(N, F)`` float32 + - Text source (``data.inputs`` + ``model.tokenizer`` set) → tokenised + ``input_ids`` of shape ``(N, T)`` long, with an ``attention_mask`` Args: cfg: Application configuration. Returns: - ``(tensor, sample_ids, input_modality)`` where ``tensor`` is the raw - data tensor (or a ragged ``list[Tensor]`` of per-image tensors for - detection), ``sample_ids`` are posix-relative file ids (``None`` for - tabular), and ``input_modality`` is the recorded - :class:`~raitap.data.types.InputModality`. + ``(tensor, sample_ids, input_modality, attention_mask)`` where + ``tensor`` is the raw data tensor (or a ragged ``list[Tensor]`` of + per-image tensors for detection), ``sample_ids`` are posix-relative + file ids (``None`` for tabular), ``input_modality`` is the recorded + :class:`~raitap.data.types.InputModality`, and ``attention_mask`` is + the tokeniser padding mask for text (``None`` for every other + modality). """ source = cfg.data.source if not source or not source.strip(): @@ -115,6 +122,21 @@ def _load_data( "Use a local path or a named sample set, e.g.: data=imagenet_samples" ) + inputs_cfg = _get_optional_config_value(cfg.data, "inputs") + model_cfg_for_text = getattr(cfg, "model", None) + tokenizer_id = _get_optional_config_value(model_cfg_for_text, "tokenizer") + if inputs_cfg is not None and tokenizer_id is not None: + from transformers import AutoTokenizer # lazy: optional dep + + from raitap.data.input_parsers.factory import create_input_parser + + parser = create_input_parser(inputs_cfg) + texts = parser.parse(source=source, sample_ids=None) + tokenizer = AutoTokenizer.from_pretrained(tokenizer_id) + input_ids, attention_mask = _tokenise_texts(texts, tokenizer=tokenizer) + sample_ids = [str(i) for i in range(len(texts))] + return input_ids, sample_ids, InputModality.text, attention_mask + # Use the run-level resolution when the orchestrator supplies it. Direct # ``Data(config)`` callers keep the legacy fallback path. if resolved_preprocessing is not None: @@ -139,7 +161,7 @@ def _load_data( # at their native resolution instead of pre-squashing them to # ``_DEMO_SIZE`` before the bundled Resize/CenterCrop sees them. tensor, sample_ids = _load_sample(source, per_image_transform=per_image_transform) - return tensor, sample_ids, InputModality.image + return tensor, sample_ids, InputModality.image, None path = get_source_path(source, kind=SourceKind.DATA) @@ -160,17 +182,24 @@ def _load_data( ragged, _resolve_sample_ids(image_files, root=path), InputModality.image, + None, ) tensor = torch.from_numpy( _stack_images_numpy(image_files, per_image_transform=per_image_transform) ) - return tensor, _resolve_sample_ids(image_files, root=path), InputModality.image + return ( + tensor, + _resolve_sample_ids(image_files, root=path), + InputModality.image, + None, + ) if tabular_files: tensor = torch.from_numpy(_concat_tabular_numpy(tabular_files)) return ( _apply_data_module_to_batch(data_module, tensor), None, InputModality.tabular, + None, ) raise FileNotFoundError( f"No supported files found in {path}.\n" @@ -182,17 +211,24 @@ def _load_data( if suffix in _IMAGE_EXTENSIONS: if is_detection: ragged = _load_images_ragged([path], per_image_transform=per_image_transform) - return ragged, _resolve_sample_ids([path], root=path.parent), InputModality.image + return ( + ragged, + _resolve_sample_ids([path], root=path.parent), + InputModality.image, + None, + ) return ( _load_images(path, per_image_transform=per_image_transform), _resolve_sample_ids([path], root=path.parent), InputModality.image, + None, ) if suffix in _TABULAR_EXTENSIONS: return ( _apply_data_module_to_batch(data_module, _load_tabular(path)), None, InputModality.tabular, + None, ) raise ValueError( @@ -237,6 +273,16 @@ def log(self, tracker: BaseTracker, **kwargs: Any) -> None: tracker.log_dataset(self.describe()) +def _tokenise_texts( + texts: list[str], *, tokenizer: Any, max_length: int = 256 +) -> tuple[torch.Tensor, torch.Tensor]: + """Tokenise raw strings into padded ``input_ids`` + ``attention_mask``.""" + enc = tokenizer( + texts, padding=True, truncation=True, max_length=max_length, return_tensors="pt" + ) + return enc["input_ids"].long(), enc["attention_mask"].long() + + def _resolve_and_parse_labels( cfg: Any, *, diff --git a/src/raitap/data/tests/test_load_text.py b/src/raitap/data/tests/test_load_text.py new file mode 100644 index 00000000..609895dc --- /dev/null +++ b/src/raitap/data/tests/test_load_text.py @@ -0,0 +1,18 @@ +import pytest + +pytest.importorskip("transformers") + +import torch +from transformers import AutoTokenizer + +from raitap.data.data import _tokenise_texts + + +def test_tokenise_texts_returns_ids_and_mask() -> None: + tok = AutoTokenizer.from_pretrained( + "hf-internal-testing/tiny-random-DistilBertForSequenceClassification" + ) + ids, mask = _tokenise_texts(["hello world", "bye"], tokenizer=tok, max_length=16) + assert ids.dtype == torch.long and ids.shape == mask.shape + assert ids.shape[0] == 2 + assert mask.sum() > 0 From 057ba4f1846727e6cd02c0677e949dbcbe407722 Mon Sep 17 00:00:00 2001 From: Stanislas Laurent Date: Sun, 5 Jul 2026 21:12:37 +0200 Subject: [PATCH 06/24] feat(deps): require text extra for tokenizer-configured models (refs #340) --- src/raitap/deps/inference.py | 30 ++++++++++++++++++++----- src/raitap/deps/tests/test_inference.py | 23 +++++++++++++++++++ 2 files changed, 48 insertions(+), 5 deletions(-) diff --git a/src/raitap/deps/inference.py b/src/raitap/deps/inference.py index 68d0baf7..7508dbcb 100644 --- a/src/raitap/deps/inference.py +++ b/src/raitap/deps/inference.py @@ -72,9 +72,9 @@ def backend_extra(model_source: str, hardware: ResolvedHardware) -> str: The extension -> extra mapping is harvested import-free from the backends' ``@register`` decorators (:func:`scan_backend_extras`): accelerator runtimes (torch/onnx) split per hardware, single-wheel ones (xgboost) do not. - Extensionless sources (built-in torchvision names, e.g. ``"resnet50"``) and - unknown extensions fall back to the torch runtime — a truly unsupported file - errors later at load time. + Extensionless sources (built-in torchvision names, e.g. ``"resnet50"``, and + HuggingFace hub ids) and unknown extensions fall back to the torch runtime — + a truly unsupported file errors later at load time. """ ext = os.path.splitext(model_source)[1].lower() spec = scan_backend_extras().get(ext) @@ -84,6 +84,26 @@ def backend_extra(model_source: str, hardware: ResolvedHardware) -> str: return _extra_for_spec(extra, supported_hardware, hardware) +def backend_extras(model: Mapping[str, Any], hardware: ResolvedHardware) -> dict[str, str]: + """Return ``{extra: origin}`` for everything a model config needs to run. + + ``backend_extra`` maps ``model.source`` to a single runtime extra by file + extension. HuggingFace text models are the one case that needs more than + that: their source is an extensionless hub id (so ``backend_extra`` falls + back to the torch runtime, which is correct — transformers runs on torch) + *and* they set ``model.tokenizer``, which additionally requires the + ``text`` extra (it carries ``transformers``). Both are returned together + so callers don't have to special-case the tokenizer signal themselves. + """ + source = model["source"] + found: dict[str, str] = {} + _add(found, backend_extra(source, hardware), f"model.source={source} + hardware={hardware}") + tokenizer = model.get("tokenizer") + if tokenizer is not None: + _add(found, "text", f"model.tokenizer={tokenizer!r}") + return found + + def _extra_for_target(target: str) -> str: cls = _class_name(target) extra = ADAPTER_EXTRAS.get(cls) or scan_adapter_extras().get(cls) @@ -175,8 +195,8 @@ def infer_extras( "(the file extension drives that choice). Add `model.source: …` " "to the config or override it via Hydra (e.g. `model.source=foo.pt`)." ) - backend = backend_extra(model["source"], hardware) - _add(extras, backend, f"model.source={model['source']} + hardware={hardware}") + for name, origin in backend_extras(model, hardware).items(): + _add(extras, name, origin) for section_key in ("transparency", "robustness", "reporting", "tracking", "metrics"): _walk_section(extras, section_key, cfg.get(section_key)) diff --git a/src/raitap/deps/tests/test_inference.py b/src/raitap/deps/tests/test_inference.py index fac0ddc6..21def043 100644 --- a/src/raitap/deps/tests/test_inference.py +++ b/src/raitap/deps/tests/test_inference.py @@ -57,6 +57,29 @@ def test_backend_extra_pure() -> None: assert backend_extra("a.ubj", ResolvedHardware.cpu) == "xgboost" +def test_tokenizer_configured_model_adds_text_extra() -> None: + # HF text models are extensionless hub-ids (e.g. a hub id with no file + # extension) with `model.tokenizer` set — deps inference must add `text` + # (which carries `transformers`) on top of the torch runtime extra that + # the extensionless fallback already infers. + cfg = { + "model": { + "source": "distilbert-base-uncased-finetuned-sst-2-english", + "tokenizer": "distilbert-base-uncased-finetuned-sst-2-english", + } + } + extras, origins = infer_extras(cfg, hardware=ResolvedHardware.cpu) + assert "text" in extras + assert "torch-cpu" in extras + assert "model.tokenizer" in origins["text"] + + +def test_no_tokenizer_does_not_add_text_extra() -> None: + cfg = {"model": {"source": "resnet50"}} + extras, _ = infer_extras(cfg, hardware=ResolvedHardware.cpu) + assert "text" not in extras + + def test_builtin_name_falls_back_to_torch() -> None: # Built-in torchvision models are extensionless (e.g. "resnet50"); they have # no backend registration, so deps inference defaults to the torch runtime. From 43ca9c5ae0a8977ad3d2e1f44dc4d4631bd22db1 Mon Sep 17 00:00:00 2001 From: Stanislas Laurent Date: Sun, 5 Jul 2026 21:17:43 +0200 Subject: [PATCH 07/24] feat(model): widen backend __call__ to accept forward kwargs (attention_mask) (refs #340) --- src/raitap/models/base_backend.py | 2 +- src/raitap/models/onnx_backend.py | 2 +- src/raitap/models/tests/test_backend.py | 2 +- .../models/tests/test_backend_kwargs.py | 22 +++++++++++++++++++ src/raitap/models/tree_backend.py | 2 +- .../phases/tests/test_forward_pass.py | 2 +- .../robustness/tests/test_e2e_real_data.py | 2 +- .../tests/test_e2e_robustness_matrix.py | 2 +- .../tests/test_e2e_statistical_sampling.py | 2 +- src/raitap/robustness/tests/test_factory.py | 2 +- .../test_assess_transparency_detection.py | 2 +- src/raitap/transparency/tests/test_factory.py | 2 +- 12 files changed, 33 insertions(+), 11 deletions(-) create mode 100644 src/raitap/models/tests/test_backend_kwargs.py diff --git a/src/raitap/models/base_backend.py b/src/raitap/models/base_backend.py index 126009aa..89212dfd 100644 --- a/src/raitap/models/base_backend.py +++ b/src/raitap/models/base_backend.py @@ -114,7 +114,7 @@ def _prepare_kwargs(self, kwargs: dict[str, Any]) -> dict[str, Any]: return kwargs @abstractmethod - def __call__(self, inputs: torch.Tensor) -> Any: + def __call__(self, inputs: torch.Tensor, **kwargs: Any) -> Any: """Run inference for ``inputs``.""" def predict_callable(self) -> Callable[[torch.Tensor], torch.Tensor]: diff --git a/src/raitap/models/onnx_backend.py b/src/raitap/models/onnx_backend.py index f5014f1a..79027e5d 100644 --- a/src/raitap/models/onnx_backend.py +++ b/src/raitap/models/onnx_backend.py @@ -155,7 +155,7 @@ def forward_numpy(self, batch: np.ndarray[Any, Any]) -> np.ndarray[Any, Any]: return np_outputs[0] return self._select_primary_numpy_output(np_outputs) - def __call__(self, inputs: torch.Tensor) -> torch.Tensor: + def __call__(self, inputs: torch.Tensor, **kwargs: Any) -> torch.Tensor: if not isinstance(inputs, torch.Tensor): raise TypeError( f"OnnxBackend expected torch.Tensor inputs, got {type(inputs).__name__}." diff --git a/src/raitap/models/tests/test_backend.py b/src/raitap/models/tests/test_backend.py index f44a0750..ad9ce4a4 100644 --- a/src/raitap/models/tests/test_backend.py +++ b/src/raitap/models/tests/test_backend.py @@ -72,7 +72,7 @@ class _B(ModelBackend): def hardware_label(self) -> str: return "test" - def __call__(self, inputs: object) -> object: + def __call__(self, inputs: object, **kwargs: object) -> object: return inputs assert _B.provides == frozenset({Capability.AUTOGRAD}) diff --git a/src/raitap/models/tests/test_backend_kwargs.py b/src/raitap/models/tests/test_backend_kwargs.py new file mode 100644 index 00000000..dc90b4ea --- /dev/null +++ b/src/raitap/models/tests/test_backend_kwargs.py @@ -0,0 +1,22 @@ +"""Coverage guard: ``ModelBackend.__call__`` admits forward kwargs (e.g. attention_mask).""" + +import torch +from torch import nn + +from raitap.models.torch_backend import TorchBackend + + +class _TwoInput(nn.Module): + def forward(self, x: torch.Tensor, attention_mask: torch.Tensor | None = None) -> torch.Tensor: + return x.sum(dim=1, keepdim=True) + (0 if attention_mask is None else attention_mask.sum()) + + +def test_backend_call_forwards_kwargs() -> None: + backend = TorchBackend(_TwoInput(), device=torch.device("cpu")) + x = torch.ones((2, 3)) + mask = torch.ones((2, 3)) + out = backend(x, attention_mask=mask) + # Assert on the VALUE, not just shape: the mask must actually reach forward. + # x.sum(dim=1)=3 per row; mask.sum()=6 => 9. Without forwarding it would be 3. + expected = x.sum(dim=1, keepdim=True) + mask.sum() + assert torch.equal(out, expected) diff --git a/src/raitap/models/tree_backend.py b/src/raitap/models/tree_backend.py index f9ddd715..128f576c 100644 --- a/src/raitap/models/tree_backend.py +++ b/src/raitap/models/tree_backend.py @@ -39,7 +39,7 @@ def fitted_estimator(self) -> Any: """Raw fitted estimator (consumed by shap.TreeExplainer).""" return self._estimator - def __call__(self, inputs: torch.Tensor) -> torch.Tensor: + def __call__(self, inputs: torch.Tensor, **kwargs: Any) -> torch.Tensor: prepared = self._prepare_inputs(inputs) array = ( prepared.detach().cpu().numpy() diff --git a/src/raitap/pipeline/phases/tests/test_forward_pass.py b/src/raitap/pipeline/phases/tests/test_forward_pass.py index fa7be4dc..e743b73b 100644 --- a/src/raitap/pipeline/phases/tests/test_forward_pass.py +++ b/src/raitap/pipeline/phases/tests/test_forward_pass.py @@ -91,7 +91,7 @@ def hardware_label(self) -> str: def _prepare_inputs(self, inputs: torch.Tensor) -> torch.Tensor: return inputs - def __call__(self, inputs: torch.Tensor) -> torch.Tensor: + def __call__(self, inputs: torch.Tensor, **kwargs: Any) -> torch.Tensor: n = inputs.shape[0] return torch.zeros(n, self.num_classes) diff --git a/src/raitap/robustness/tests/test_e2e_real_data.py b/src/raitap/robustness/tests/test_e2e_real_data.py index f945e05d..6b09f373 100644 --- a/src/raitap/robustness/tests/test_e2e_real_data.py +++ b/src/raitap/robustness/tests/test_e2e_real_data.py @@ -55,7 +55,7 @@ def _prepare_inputs(self, inputs: torch.Tensor) -> torch.Tensor: def _prepare_kwargs(self, kwargs: dict[str, Any]) -> dict[str, Any]: return kwargs - def __call__(self, inputs: torch.Tensor) -> torch.Tensor: + def __call__(self, inputs: torch.Tensor, **kwargs: Any) -> torch.Tensor: return self._model(inputs) def autograd_module(self) -> torch.nn.Module: diff --git a/src/raitap/robustness/tests/test_e2e_robustness_matrix.py b/src/raitap/robustness/tests/test_e2e_robustness_matrix.py index 0c63899f..6b53fd3b 100644 --- a/src/raitap/robustness/tests/test_e2e_robustness_matrix.py +++ b/src/raitap/robustness/tests/test_e2e_robustness_matrix.py @@ -51,7 +51,7 @@ def _prepare_inputs(self, inputs: torch.Tensor) -> torch.Tensor: def _prepare_kwargs(self, kwargs: dict[str, Any]) -> dict[str, Any]: return kwargs - def __call__(self, inputs: torch.Tensor) -> torch.Tensor: + def __call__(self, inputs: torch.Tensor, **kwargs: Any) -> torch.Tensor: return self._model(inputs) def autograd_module(self) -> torch.nn.Module: diff --git a/src/raitap/robustness/tests/test_e2e_statistical_sampling.py b/src/raitap/robustness/tests/test_e2e_statistical_sampling.py index 2cac5368..5fbff36b 100644 --- a/src/raitap/robustness/tests/test_e2e_statistical_sampling.py +++ b/src/raitap/robustness/tests/test_e2e_statistical_sampling.py @@ -42,7 +42,7 @@ def _prepare_inputs(self, inputs: torch.Tensor) -> torch.Tensor: def _prepare_kwargs(self, kwargs: dict[str, Any]) -> dict[str, Any]: return kwargs - def __call__(self, inputs: torch.Tensor) -> torch.Tensor: + def __call__(self, inputs: torch.Tensor, **kwargs: Any) -> torch.Tensor: return self._model(inputs) def autograd_module(self) -> torch.nn.Module: diff --git a/src/raitap/robustness/tests/test_factory.py b/src/raitap/robustness/tests/test_factory.py index 8366acb4..c48e7971 100644 --- a/src/raitap/robustness/tests/test_factory.py +++ b/src/raitap/robustness/tests/test_factory.py @@ -45,7 +45,7 @@ def _prepare_inputs(self, inputs: torch.Tensor) -> torch.Tensor: def _prepare_kwargs(self, kwargs: dict[str, Any]) -> dict[str, Any]: return kwargs - def __call__(self, inputs: torch.Tensor) -> torch.Tensor: + def __call__(self, inputs: torch.Tensor, **kwargs: Any) -> torch.Tensor: return self._model(inputs) def autograd_module(self) -> torch.nn.Module: diff --git a/src/raitap/transparency/tests/test_assess_transparency_detection.py b/src/raitap/transparency/tests/test_assess_transparency_detection.py index b10bacef..1979ae90 100644 --- a/src/raitap/transparency/tests/test_assess_transparency_detection.py +++ b/src/raitap/transparency/tests/test_assess_transparency_detection.py @@ -52,7 +52,7 @@ def hardware_label(self) -> str: def _prepare_kwargs(self, kwargs: dict[str, Any]) -> dict[str, Any]: return kwargs - def __call__(self, inputs: torch.Tensor) -> Any: + def __call__(self, inputs: torch.Tensor, **kwargs: Any) -> Any: raise NotImplementedError("routing test mocks the explain path") diff --git a/src/raitap/transparency/tests/test_factory.py b/src/raitap/transparency/tests/test_factory.py index 9cb9488b..2cdfbdb3 100644 --- a/src/raitap/transparency/tests/test_factory.py +++ b/src/raitap/transparency/tests/test_factory.py @@ -72,7 +72,7 @@ def _prepare_inputs(self, inputs: torch.Tensor) -> torch.Tensor: def _prepare_kwargs(self, kwargs: dict[str, Any]) -> dict[str, Any]: return kwargs - def __call__(self, inputs: torch.Tensor) -> torch.Tensor: + def __call__(self, inputs: torch.Tensor, **kwargs: Any) -> torch.Tensor: return self._model(inputs) def autograd_module(self) -> torch.nn.Module: From 84d28eeeb742cd5516fc2c560b6f3e52771e3ee8 Mon Sep 17 00:00:00 2001 From: Stanislas Laurent Date: Sun, 5 Jul 2026 21:28:32 +0200 Subject: [PATCH 08/24] feat(model): thread attention_mask through the classification forward pass (refs #340) --- src/raitap/pipeline/orchestrator.py | 8 ++++- src/raitap/pipeline/phases/forward_pass.py | 16 ++++++--- src/raitap/task_families/classification.py | 24 +++++++------ .../tests/test_classification_forward_mask.py | 34 +++++++++++++++++++ 4 files changed, 66 insertions(+), 16 deletions(-) create mode 100644 src/raitap/task_families/tests/test_classification_forward_mask.py diff --git a/src/raitap/pipeline/orchestrator.py b/src/raitap/pipeline/orchestrator.py index 1c9ec656..bce4b20a 100644 --- a/src/raitap/pipeline/orchestrator.py +++ b/src/raitap/pipeline/orchestrator.py @@ -170,8 +170,14 @@ def run_without_tracking( # Logged from the orchestrator but logically a model operation (same pattern # as the reporting log below) so the chip reads "Models", not blank/infra. raitap_log.info("Running model forward pass...", module=Module.models) + attention_mask = getattr(data, "attention_mask", None) with torch.no_grad(): - forward_output = forward_pass(config, model.backend, data.tensor) + forward_output = forward_pass( + config, + model.backend, + data.tensor, + forward_kwargs=None if attention_mask is None else {"attention_mask": attention_mask}, + ) context = PhaseContext( config=config, diff --git a/src/raitap/pipeline/phases/forward_pass.py b/src/raitap/pipeline/phases/forward_pass.py index 2062f319..90a99786 100644 --- a/src/raitap/pipeline/phases/forward_pass.py +++ b/src/raitap/pipeline/phases/forward_pass.py @@ -118,7 +118,11 @@ def _validate_inputs_for_task(inputs: torch.Tensor | DetectionInputs, task_kind: def forward_pass( - config: AppConfig, backend: Any, inputs: torch.Tensor | DetectionInputs + config: AppConfig, + backend: Any, + inputs: torch.Tensor | DetectionInputs, + *, + forward_kwargs: dict[str, torch.Tensor] | None = None, ) -> ForwardOutput: """Run the model backend forward in chunks of ``forward_batch_size``. @@ -131,14 +135,18 @@ def forward_pass( native-resolution ``(C, H, W)`` tensor per image (sizes may differ, so they cannot be stacked). For classification, ``inputs`` is a dense ``(N, C, H, W)`` tensor. + + ``forward_kwargs`` carries extra per-sample forward args (e.g. a text + model's ``attention_mask``) into ``ForwardContext.extras``, from which the + resolved task family slices and forwards them. ``None``/empty leaves the + image/tabular path byte-identical to a plain forward. """ batch_size = resolve_forward_batch_size(config) task_kind = backend.task_kind _validate_inputs_for_task(inputs, task_kind) family = resolve_task_family(task_kind) - payload = family.extract_forward( - ForwardContext(backend=backend, inputs=inputs), batch_size=batch_size - ) + ctx = ForwardContext(backend=backend, inputs=inputs, extras=dict(forward_kwargs or {})) + payload = family.extract_forward(ctx, batch_size=batch_size) output_kind = ( OutputKind.PROBABILITIES if Capability.PREDICT_PROBA in getattr(backend, "provides", frozenset()) diff --git a/src/raitap/task_families/classification.py b/src/raitap/task_families/classification.py index 3f8c1bf1..19f48645 100644 --- a/src/raitap/task_families/classification.py +++ b/src/raitap/task_families/classification.py @@ -76,22 +76,24 @@ def extract_forward(self, ctx: ForwardContext, *, batch_size: int) -> torch.Tens from raitap.pipeline.phases.forward_pass import extract_primary_tensor backend, inputs = ctx.backend, ctx.inputs + mask = ctx.extras.get("attention_mask") total_batch = len(inputs) - if total_batch <= batch_size: - prepared = backend._prepare_inputs(inputs) - out = extract_primary_tensor(backend(prepared)).detach().cpu() + + def _call(lo: int, hi: int) -> torch.Tensor: + prepared = backend._prepare_inputs(inputs[lo:hi]) + kwargs = {} if mask is None else {"attention_mask": mask[lo:hi]} + out = extract_primary_tensor(backend(prepared, **kwargs)).detach().cpu() del prepared if torch.cuda.is_available(): torch.cuda.empty_cache() return out - chunks: list[torch.Tensor] = [] - for start in range(0, total_batch, batch_size): - end = min(start + batch_size, total_batch) - prepared = backend._prepare_inputs(inputs[start:end]) - chunks.append(extract_primary_tensor(backend(prepared)).detach().cpu()) - del prepared - if torch.cuda.is_available(): - torch.cuda.empty_cache() + + if total_batch <= batch_size: + return _call(0, total_batch) + chunks = [ + _call(start, min(start + batch_size, total_batch)) + for start in range(0, total_batch, batch_size) + ] return torch.cat(chunks, dim=0) def payload_batch_size(self, payload: Any) -> int: diff --git a/src/raitap/task_families/tests/test_classification_forward_mask.py b/src/raitap/task_families/tests/test_classification_forward_mask.py new file mode 100644 index 00000000..cde63fa4 --- /dev/null +++ b/src/raitap/task_families/tests/test_classification_forward_mask.py @@ -0,0 +1,34 @@ +"""RED->GREEN regression: attention_mask threads through forward_pass (#340 T7).""" + +from __future__ import annotations + +from types import SimpleNamespace +from typing import Any + +import torch +from torch import nn + +from raitap.models.torch_backend import TorchBackend +from raitap.pipeline.phases.forward_pass import forward_pass + + +class _MaskAware(nn.Module): + def forward( + self, ids: torch.Tensor, attention_mask: torch.Tensor | None = None + ) -> torch.Tensor: + assert attention_mask is not None, "mask must reach forward" + return ids.float() @ torch.ones((ids.shape[1], 2)) + + +def _config() -> Any: # a fake AppConfig-shaped object for forward_pass + return SimpleNamespace(run=SimpleNamespace(forward_batch_size=2), data=None) + + +def test_mask_threads_to_forward() -> None: + backend = TorchBackend(_MaskAware(), device=torch.device("cpu")) + ids = torch.ones((5, 4), dtype=torch.long) + mask = torch.ones((5, 4), dtype=torch.long) + out = forward_pass(_config(), backend, ids, forward_kwargs={"attention_mask": mask}) + # ForwardOutput has no `.predictions`; the classification payload is the + # logits tensor itself (see `ForwardOutput.payload` / `.as_classification()`). + assert out.payload.shape[0] == 5 From acff61f04cbd2e397793a0af933ca27ff13f4fba Mon Sep 17 00:00:00 2001 From: Stanislas Laurent Date: Sun, 5 Jul 2026 21:28:33 +0200 Subject: [PATCH 09/24] feat(transparency): pass attention_mask as Captum additional_forward_args (refs #340) --- .../explainers/captum_explainer.py | 19 ++- .../tests/test_captum_text_attribution.py | 116 ++++++++++++++++++ 2 files changed, 133 insertions(+), 2 deletions(-) create mode 100644 src/raitap/transparency/tests/test_captum_text_attribution.py diff --git a/src/raitap/transparency/explainers/captum_explainer.py b/src/raitap/transparency/explainers/captum_explainer.py index edf6dd27..44edeb0b 100644 --- a/src/raitap/transparency/explainers/captum_explainer.py +++ b/src/raitap/transparency/explainers/captum_explainer.py @@ -237,6 +237,15 @@ def _default_captum_invoker(ctx: AttributionInvokeCtx) -> torch.Tensor: call_kwargs = dict(ctx.call_kwargs) target = call_kwargs.pop("target", None) baselines = call_kwargs.pop("baselines", None) + # ``attention_mask`` (e.g. from a tokenized text input) is not attributed or + # perturbed like a regular input feature; Captum passes it through untouched + # to the forward as a positional ``additional_forward_args`` entry (#340). + additional_forward_args = call_kwargs.pop("attention_mask", None) + extra_call_kwargs = ( + {"additional_forward_args": (additional_forward_args,)} + if additional_forward_args is not None + else {} + ) try: method_class = getattr(captum_attr, explainer.algorithm) @@ -259,8 +268,14 @@ def _default_captum_invoker(ctx: AttributionInvokeCtx) -> torch.Tensor: with explainer._rethrow(): method = method_class(ctx.model, **init_kwargs) if baselines is not None: - return method.attribute(ctx.inputs, target=target, baselines=baselines, **call_kwargs) - return method.attribute(ctx.inputs, target=target, **call_kwargs) + return method.attribute( + ctx.inputs, + target=target, + baselines=baselines, + **extra_call_kwargs, + **call_kwargs, + ) + return method.attribute(ctx.inputs, target=target, **extra_call_kwargs, **call_kwargs) def _needs_layer_resolution(algorithm: str) -> bool: diff --git a/src/raitap/transparency/tests/test_captum_text_attribution.py b/src/raitap/transparency/tests/test_captum_text_attribution.py new file mode 100644 index 00000000..b3e7e242 --- /dev/null +++ b/src/raitap/transparency/tests/test_captum_text_attribution.py @@ -0,0 +1,116 @@ +"""Captum text attribution: ``additional_forward_args`` + ``LayerIntegratedGradients`` (#340). + +Two layers of coverage: + +- ``test_layer_ig_produces_per_token_attributions`` exercises captum directly + against the tiny HF backend to lock down the mechanism (token ids are + discrete, so attribution must run over the embedding layer via + ``LayerIntegratedGradients``, with the attention mask riding as a + non-attributed ``additional_forward_args`` entry). +- ``test_default_invoker_pops_attention_mask_into_additional_forward_args`` + verifies raitap's own dispatch: ``_default_captum_invoker`` pops a + designated ``attention_mask`` call kwarg and forwards it as + ``additional_forward_args`` instead of double-passing it as a normal + ``attribute()`` kwarg. +""" + +from __future__ import annotations + +from typing import Any, ClassVar, cast + +import pytest +import torch + +from raitap.transparency.explainers.base_explainer import AttributionInvokeCtx +from raitap.transparency.explainers.captum_explainer import CaptumExplainer, _default_captum_invoker + + +def test_layer_ig_produces_per_token_attributions() -> None: + pytest.importorskip("transformers") + pytest.importorskip("captum") + + from raitap.models.torch_backend import load_hf_text_backend + + backend = load_hf_text_backend( + "hf-internal-testing/tiny-random-DistilBertForSequenceClassification", + tokenizer="hf-internal-testing/tiny-random-DistilBertForSequenceClassification", + device=torch.device("cpu"), + ) + enc = backend.tokenizer(["hello world"], padding=True, return_tensors="pt") + ids, mask = enc["input_ids"], enc["attention_mask"] + + # Attribution via LayerIntegratedGradients over the embedding layer, mask as + # additional_forward_args (token ids are discrete -> no gradient on input_ids). + from captum.attr import LayerIntegratedGradients + + emb = cast("Any", backend.model).get_input_embeddings() + lig = LayerIntegratedGradients(lambda i, m: backend.model(i, attention_mask=m).logits, emb) + attr = cast("torch.Tensor", lig.attribute(ids, target=0, additional_forward_args=(mask,))) + per_token = attr.sum(dim=-1) + assert per_token.shape == ids.shape + + +class _RecordingMethod: + """Fake captum method: records the kwargs ``.attribute()`` is called with.""" + + last_call_kwargs: ClassVar[dict[str, object]] = {} + + def __init__(self, model: object, **init_kwargs: object) -> None: + self.model = model + self.init_kwargs = init_kwargs + + def attribute(self, inputs: torch.Tensor, **kwargs: object) -> torch.Tensor: + _RecordingMethod.last_call_kwargs = kwargs + return torch.zeros_like(inputs, dtype=torch.float32) + + +class _FakeCaptumAttr: + IntegratedGradients = _RecordingMethod + + +def test_default_invoker_pops_attention_mask_into_additional_forward_args() -> None: + explainer = CaptumExplainer(algorithm="IntegratedGradients") + model = torch.nn.Linear(4, 2) + inputs = torch.randn(2, 4) + mask = torch.ones(2, 4) + + ctx = AttributionInvokeCtx( + explainer=explainer, + library=_FakeCaptumAttr, + model=model, + inputs=inputs, + input_spec=None, + call_kwargs={"target": 0, "baselines": None, "attention_mask": mask}, + ) + + _default_captum_invoker(ctx) + + recorded = _RecordingMethod.last_call_kwargs + assert "attention_mask" not in recorded + additional_forward_args = recorded["additional_forward_args"] + assert isinstance(additional_forward_args, tuple) + assert len(additional_forward_args) == 1 + assert additional_forward_args[0] is mask + assert recorded["target"] == 0 + + +def test_default_invoker_omits_additional_forward_args_when_no_mask() -> None: + """Existing image/tabular paths (no ``attention_mask`` kwarg) stay unchanged.""" + explainer = CaptumExplainer(algorithm="IntegratedGradients") + model = torch.nn.Linear(4, 2) + inputs = torch.randn(2, 4) + + ctx = AttributionInvokeCtx( + explainer=explainer, + library=_FakeCaptumAttr, + model=model, + inputs=inputs, + input_spec=None, + call_kwargs={"target": 0, "baselines": None}, + ) + + _default_captum_invoker(ctx) + + recorded = _RecordingMethod.last_call_kwargs + assert "additional_forward_args" not in recorded + assert "attention_mask" not in recorded From b70085658083176082e2add1d33a515df6248745 Mon Sep 17 00:00:00 2001 From: Stanislas Laurent Date: Sun, 5 Jul 2026 22:00:15 +0200 Subject: [PATCH 10/24] feat(transparency): render text attribution end-to-end (mask to explainer, LayerIG token reduction, batch panels) (refs #340) --- docs/modules/transparency/visualisers.md | 4 +- src/raitap/task_families/classification.py | 14 ++++- .../explainers/captum_explainer.py | 57 +++++++++++++++++-- src/raitap/transparency/semantics.py | 21 ++++--- .../visualisers/captum_visualisers.py | 46 ++++++++++----- .../visualisers/tests/test_visualisers.py | 19 ++++--- 6 files changed, 124 insertions(+), 37 deletions(-) diff --git a/docs/modules/transparency/visualisers.md b/docs/modules/transparency/visualisers.md index df4e5f35..29898d47 100644 --- a/docs/modules/transparency/visualisers.md +++ b/docs/modules/transparency/visualisers.md @@ -87,11 +87,11 @@ the factory rejects mismatches at YAML parse time. See :registry: captum_text :wraps: a lightweight Matplotlib renderer (Captum's native text visualiser emits HTML rather than a Matplotlib figure) :intro: Horizontal bar chart of per-token attribution scores. Positive contributions render in warm red, negative in cool blue, so you can read off which tokens pushed the prediction in which direction at a glance. -:how-to-read: One horizontal bar per token, in token order top-to-bottom; bar length is the token's attribution magnitude, warm red means it pushed the prediction up and cool blue means down. Scan it to see which words drove the call. +:how-to-read: One horizontal bar per token, in token order top-to-bottom; bar length is the token's attribution magnitude, warm red means it pushed the prediction up and cool blue means down. Scan it to see which words drove the call. A batch renders one bar panel per sample. :kwarg: token_labels :default: `None` :meaning: Per-token strings used as y-axis labels. Falls back to `tok_0 … tok_N` when omitted. -:compat: Scope: `LOCAL`. Output space: `TOKEN_SEQUENCE`. Requires text input metadata and a 1-D token attribution tensor with the `TOKENS` / `TOKEN_SEQUENCE` layout. +:compat: Scope: `LOCAL`. Output space: `TOKEN_SEQUENCE`. Requires text input metadata and a token attribution tensor with the `TOKENS` / `TOKEN_SEQUENCE` layout: 1-D `(T,)` for a single sequence or 2-D `(B, T)` for a batch. ::::: ## SHAP diff --git a/src/raitap/task_families/classification.py b/src/raitap/task_families/classification.py index 19f48645..fdfeba64 100644 --- a/src/raitap/task_families/classification.py +++ b/src/raitap/task_families/classification.py @@ -126,6 +126,18 @@ def explain(self, ctx: ExplainContext) -> list: # device-moved exactly once, matching the pre-refactor behaviour. runtime_kwargs = prepared.backend._prepare_kwargs(runtime_kwargs) + # Text inputs carry a per-sample ``attention_mask`` alongside the token + # ids. It is not attributed like an input feature; thread it through as a + # call-kwarg so the explainer's invoker forwards it to the model (the + # Captum invoker pops it into ``additional_forward_args``). + # ``_slice_kwargs_for_batch`` slices it per batch (leading dim == batch + # size). Absent for image/tabular inputs, so those paths stay + # byte-identical (no extra kwarg is added). (#340) + mask_kwargs: dict[str, Any] = {} + attention_mask = getattr(ctx.data, "attention_mask", None) + if attention_mask is not None: + mask_kwargs = prepared.backend._prepare_kwargs({"attention_mask": attention_mask}) + resolved_sample_names = prepared.raitap_kwargs.get("sample_names") if resolved_sample_names is not None: resolved_list = list(resolved_sample_names) @@ -147,7 +159,7 @@ def explain(self, ctx: ExplainContext) -> list: visualisers=prepared.visualisers, raitap_kwargs=prepared.raitap_kwargs, call_provenance=prepared.call_provenance, - **{**prepared.merged_kwargs, **runtime_kwargs}, + **{**prepared.merged_kwargs, **runtime_kwargs, **mask_kwargs}, ) result._visualise() # populates result.visualisations (was run_adapters) return [result] diff --git a/src/raitap/transparency/explainers/captum_explainer.py b/src/raitap/transparency/explainers/captum_explainer.py index 44edeb0b..dcae3321 100644 --- a/src/raitap/transparency/explainers/captum_explainer.py +++ b/src/raitap/transparency/explainers/captum_explainer.py @@ -2,12 +2,13 @@ from __future__ import annotations -from typing import TYPE_CHECKING, cast +from typing import TYPE_CHECKING, Any, cast from raitap.transparency.contracts import ( BaselineCardinality, BaselineMode, ExplainerAlgorithmSpec, + InputKind, MethodFamily, StructuredOutputSpec, StructuredPayloadKind, @@ -265,17 +266,44 @@ def _default_captum_invoker(ctx: AttributionInvokeCtx) -> torch.Tensor: if explainer.algorithm == "Occlusion": call_kwargs = _normalise_occlusion_kwargs(call_kwargs) + # Text models (HF ``AutoModelForSequenceClassification``) return a + # ``SequenceClassifierOutput`` object, not a bare tensor; Captum needs the + # forward to yield a tensor. Wrap the forward to reduce it to logits. The + # ``layer`` above is resolved on the real ``ctx.model`` so LayerIntegratedGradients + # hooks fire on the same module the wrapper calls. Image/tabular inputs pass + # ``ctx.model`` through unchanged (byte-identical). (#340) + forward_func = _forward_func_for(ctx) + with explainer._rethrow(): - method = method_class(ctx.model, **init_kwargs) + method = method_class(forward_func, **init_kwargs) if baselines is not None: - return method.attribute( + attributions = method.attribute( ctx.inputs, target=target, baselines=baselines, **extra_call_kwargs, **call_kwargs, ) - return method.attribute(ctx.inputs, target=target, **extra_call_kwargs, **call_kwargs) + else: + attributions = method.attribute( + ctx.inputs, target=target, **extra_call_kwargs, **call_kwargs + ) + return _reduce_text_embedding_dim(attributions, ctx) + + +def _reduce_text_embedding_dim(attributions: Any, ctx: AttributionInvokeCtx) -> Any: + """Collapse the embedding axis of text attributions to per-token scores. + + Layer attribution over a text embedding layer returns ``(B, T, H)`` (one + vector per token). The canonical Captum text convention sums over the + embedding dimension ``H`` to get a single ``(B, T)`` importance per token, + which the ``TOKEN_SEQUENCE`` output space + ``CaptumTextVisualiser`` consume. + Non-text inputs (and already-reduced outputs) are returned unchanged. (#340) + """ + kind = getattr(ctx.input_spec, "kind", None) + if kind is InputKind.TEXT and getattr(attributions, "ndim", 0) == 3: + return attributions.sum(dim=-1) + return attributions def _needs_layer_resolution(algorithm: str) -> bool: @@ -298,6 +326,27 @@ def _resolve_layer(model: nn.Module, layer_path: str) -> nn.Module: return layer +def _forward_func_for(ctx: AttributionInvokeCtx) -> Any: + """Return the callable Captum should attribute through. + + Text models emit an object output (HF ``SequenceClassifierOutput``); Captum + requires a tensor, so wrap the forward to reduce it to the primary tensor + (logits). Non-text inputs return ``ctx.model`` unchanged so the image/tabular + paths stay byte-identical. (#340) + """ + model = ctx.model + kind = getattr(ctx.input_spec, "kind", None) + if kind is not InputKind.TEXT: + return model + + from raitap.pipeline.phases.forward_pass import extract_primary_tensor + + def _logits_forward(*forward_args: Any) -> Any: + return extract_primary_tensor(model(*forward_args)) + + return _logits_forward + + def _normalise_occlusion_kwargs(attr_kwargs: dict[str, object]) -> dict[str, object]: normalised = dict(attr_kwargs) diff --git a/src/raitap/transparency/semantics.py b/src/raitap/transparency/semantics.py index 92f2f368..e3688da5 100644 --- a/src/raitap/transparency/semantics.py +++ b/src/raitap/transparency/semantics.py @@ -255,24 +255,29 @@ def infer_output_space( requires_interpolation=False, ) - if layer_path is not None: - # Non-CAM Layer* attribution is layer-space, not input-aligned; tag - # LAYER_ACTIVATION (skips INPUT_FEATURES shape validation). (#267) + # Text token attribution is token-aligned even via a Layer* method + # (LayerIntegratedGradients over the embeddings is the canonical text path): + # the embedding axis is collapsed to a per-token score, so the result lives + # in TOKEN_SEQUENCE space, not opaque layer space. Text therefore takes + # precedence over the generic ``layer_path`` branch below. (#340) + if input_kind is InputKind.TEXT or input_layout is TensorLayout.TOKEN_SEQUENCE: return OutputSpaceSpec( - space=ExplanationOutputSpace.LAYER_ACTIVATION, + space=ExplanationOutputSpace.TOKEN_SEQUENCE, shape=shape, layout=input_spec.layout, - layer_path=layer_path, feature_names=features, - requires_interpolation=False, ) - if input_kind is InputKind.TEXT or input_layout is TensorLayout.TOKEN_SEQUENCE: + if layer_path is not None: + # Non-CAM Layer* attribution is layer-space, not input-aligned; tag + # LAYER_ACTIVATION (skips INPUT_FEATURES shape validation). (#267) return OutputSpaceSpec( - space=ExplanationOutputSpace.TOKEN_SEQUENCE, + space=ExplanationOutputSpace.LAYER_ACTIVATION, shape=shape, layout=input_spec.layout, + layer_path=layer_path, feature_names=features, + requires_interpolation=False, ) if input_kind is InputKind.TIME_SERIES or input_layout is TensorLayout.BATCH_TIME_CHANNEL: diff --git a/src/raitap/transparency/visualisers/captum_visualisers.py b/src/raitap/transparency/visualisers/captum_visualisers.py index 102490c8..8a0bb253 100644 --- a/src/raitap/transparency/visualisers/captum_visualisers.py +++ b/src/raitap/transparency/visualisers/captum_visualisers.py @@ -247,7 +247,9 @@ def _has_token_layout(explanation: object, attributions: object) -> bool: if any(layout and layout not in valid_layouts for layout in layouts): return False shape = _output_shape(explanation, attributions) - return shape is not None and len(shape) == 1 + # 1-D ``(T,)`` for a single token sequence, or 2-D ``(B, T)`` for a batch of + # per-token scores (one row per sample) — both are token-sequence layouts. + return shape is not None and len(shape) in {1, 2} @transparency_visualiser( @@ -661,21 +663,35 @@ def visualise( Returns: Matplotlib Figure with a horizontal bar chart of token importance. + A 2-D ``(B, T)`` batch renders one bar panel per sample (row). """ - attr = _to_numpy(attributions).ravel() - n = len(attr) - labels = token_labels if token_labels else [f"tok_{i}" for i in range(n)] - - fig, ax = plt.subplots(figsize=(8, max(4, n * 0.35))) - colors = ["#e74c3c" if v > 0 else "#3498db" for v in attr] - y = np.arange(n) - ax.barh(y, attr, color=colors) - ax.set_yticks(y) - ax.set_yticklabels(labels, fontsize=8) - ax.axvline(0, color="black", linewidth=0.8) - ax.set_xlabel("Attribution") - ax.set_title("Token Attribution") - ax.grid(axis="x", alpha=0.3) + attr = _to_numpy(attributions) + # Normalise to a batch of rows: 1-D ``(T,)`` is a single sample. + rows = attr.reshape(1, -1) if attr.ndim == 1 else attr.reshape(attr.shape[0], -1) + n_samples, n_tokens = rows.shape + + fig, axes = plt.subplots( + n_samples, + 1, + figsize=(8, max(4, n_tokens * 0.35) * n_samples), + squeeze=False, + ) + y = np.arange(n_tokens) + labels = token_labels if token_labels else [f"tok_{i}" for i in range(n_tokens)] + for row_index in range(n_samples): + ax = axes[row_index][0] + values = rows[row_index] + colors = ["#e74c3c" if v > 0 else "#3498db" for v in values] + ax.barh(y, values, color=colors) + ax.set_yticks(y) + ax.set_yticklabels(labels, fontsize=8) + ax.axvline(0, color="black", linewidth=0.8) + ax.set_xlabel("Attribution") + title = ( + "Token Attribution" if n_samples == 1 else f"Token Attribution (sample {row_index})" + ) + ax.set_title(title) + ax.grid(axis="x", alpha=0.3) fig.tight_layout() return fig diff --git a/src/raitap/transparency/visualisers/tests/test_visualisers.py b/src/raitap/transparency/visualisers/tests/test_visualisers.py index 5739b481..7e5dd5bb 100644 --- a/src/raitap/transparency/visualisers/tests/test_visualisers.py +++ b/src/raitap/transparency/visualisers/tests/test_visualisers.py @@ -759,13 +759,6 @@ def test_validate_explanation_accepts_token_sequence_metadata(self) -> None: @pytest.mark.parametrize( "explanation", [ - _explanation( - input_kind="text", - input_layout="TOKENS", - output_layout="TOKENS", - output_space=ExplanationOutputSpace.TOKEN_SEQUENCE, - shape=(2, 12), - ), _explanation( input_kind="text", input_layout="B,F", @@ -786,6 +779,18 @@ def test_validate_explanation_rejects_incompatible_text_layout_or_shape( None, ) + def test_validate_explanation_accepts_batched_token_sequence(self) -> None: + # A 2-D ``(B, T)`` batch of per-token scores is a valid token-sequence + # layout (one row per sample), rendered as one bar panel per sample. (#340) + explanation = _explanation( + input_kind="text", + input_layout="TOKENS", + output_layout="TOKENS", + output_space=ExplanationOutputSpace.TOKEN_SEQUENCE, + shape=(2, 12), + ) + CaptumTextVisualiser().validate_explanation(explanation, torch.zeros(2, 12), None) + @pytest.mark.parametrize( "method_families", [ From 8b901e2f315a5864d9b5c4fdb19a9992492319ba Mon Sep 17 00:00:00 2001 From: Stanislas Laurent Date: Sun, 5 Jul 2026 22:00:16 +0200 Subject: [PATCH 11/24] feat(config): add text-classification SST-2 end-to-end use case (refs #340) --- .../text-classification-sst2/README.md | 53 ++++++++++++ .../text-classification-sst2/assessment.yaml | 71 ++++++++++++++++ .../text-classification-sst2/build_data.py | 57 +++++++++++++ .../tests/test_text_classification_e2e.py | 83 +++++++++++++++++++ 4 files changed, 264 insertions(+) create mode 100644 contributor-configs/text-classification-sst2/README.md create mode 100644 contributor-configs/text-classification-sst2/assessment.yaml create mode 100644 contributor-configs/text-classification-sst2/build_data.py create mode 100644 src/raitap/pipeline/tests/test_text_classification_e2e.py diff --git a/contributor-configs/text-classification-sst2/README.md b/contributor-configs/text-classification-sst2/README.md new file mode 100644 index 00000000..56faac2a --- /dev/null +++ b/contributor-configs/text-classification-sst2/README.md @@ -0,0 +1,53 @@ +# Text-classification (SST-2 sentiment) smoke config + +End-to-end check for the text input modality (#340): loads a HuggingFace +sentiment model + tokenizer, tokenises a small text CSV into `input_ids` + +`attention_mask`, runs the forward pass + metrics, then `LayerIntegratedGradients` +token attribution over the embedding layer and renders per-token importances +with `CaptumTextVisualiser`. + +This is the keystone that proves the text path works: load -> tokenise -> +predict -> token attribution -> HTML report with per-token panels. + +## Run + +Artifacts (`artifacts/reviews.csv`, `artifacts/labels.csv`) are gitignored +(`contributor-configs/*/artifacts/`), so generate them first: + +```bash +uv run --extra text python contributor-configs/text-classification-sst2/build_data.py +``` + +Then run the assessment: + +```bash +uv run raitap --config-dir contributor-configs/text-classification-sst2 --config-name assessment +``` + +`raitap-deps` infers the `text` (transformers), `captum`, `metrics`, and `html` +extras and installs them. The model +(`distilbert-base-uncased-finetuned-sst-2-english`, ~270 MB) downloads from the +HuggingFace hub on first run. The HTML report lands under +`outputs//