dataeval-flow 0.2.0__tar.gz → 0.2.1__tar.gz

{dataeval_flow-0.2.0 → dataeval_flow-0.2.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: dataeval-flow
-Version: 0.2.0
+Version: 0.2.1
 Summary: DataEval Workflows container for data evaluation
 Project-URL: Homepage, https://github.com/aria-ml/dataeval-flow
 Project-URL: Documentation, https://dataeval-flow.readthedocs.io/
@@ -29,25 +29,31 @@ Requires-Dist: textual>=3.0; extra == 'app'
 Provides-Extra: cpu
 Requires-Dist: torch>=2.2.0; extra == 'cpu'
 Requires-Dist: torchvision>=0.17.0; extra == 'cpu'
-Provides-Extra: cu118
-Requires-Dist: torch>=2.2.0; extra == 'cu118'
-Requires-Dist: torchvision>=0.17.0; extra == 'cu118'
-Provides-Extra: cu128
-Requires-Dist: torch>=2.2.0; extra == 'cu128'
-Requires-Dist: torchvision>=0.17.0; extra == 'cu128'
+Provides-Extra: cu126
+Requires-Dist: torch>=2.2.0; extra == 'cu126'
+Requires-Dist: torchvision>=0.17.0; extra == 'cu126'
+Provides-Extra: cu130
+Requires-Dist: torch>=2.2.0; extra == 'cu130'
+Requires-Dist: torchvision>=0.17.0; extra == 'cu130'
 Provides-Extra: onnx
 Requires-Dist: onnx>=1.15; extra == 'onnx'
 Requires-Dist: onnxruntime<1.24,>=1.20; (python_version == '3.10') and extra == 'onnx'
 Requires-Dist: onnxruntime>=1.20; (python_version >= '3.11') and extra == 'onnx'
-Provides-Extra: onnx-gpu
-Requires-Dist: onnx>=1.15; extra == 'onnx-gpu'
-Requires-Dist: onnxruntime-gpu<1.24,>=1.23.2; (python_version == '3.10') and extra == 'onnx-gpu'
-Requires-Dist: onnxruntime-gpu<1.27,>=1.23.2; (python_version >= '3.11') and extra == 'onnx-gpu'
-Requires-Dist: onnxruntime-gpu<1.27,>=1.24; (python_version >= '3.14') and extra == 'onnx-gpu'
+Provides-Extra: onnx-cu126
+Requires-Dist: onnx>=1.15; extra == 'onnx-cu126'
+Requires-Dist: onnxruntime-gpu<1.24,>=1.20; (python_version == '3.10' and extra != 'onnx-cu130') and extra == 'onnx-cu126'
+Requires-Dist: onnxruntime-gpu<1.27,>=1.20; (python_version >= '3.11' and extra != 'onnx-cu130') and extra == 'onnx-cu126'
+Requires-Dist: onnxruntime-gpu<1.27,>=1.24; (python_version >= '3.14' and extra != 'onnx-cu130') and extra == 'onnx-cu126'
+Provides-Extra: onnx-cu130
+Requires-Dist: onnx>=1.15; extra == 'onnx-cu130'
+Requires-Dist: onnxruntime-gpu>=1.27; (python_version >= '3.11' and extra != 'onnx-cu126') and extra == 'onnx-cu130'
+Requires-Dist: onnxruntime>=1.20; (python_version == '3.10' and extra != 'onnx-cu126') and extra == 'onnx-cu130'
 Provides-Extra: ontology
 Requires-Dist: dataeval[ontology]==1.1.0-rc6; extra == 'ontology'
 Provides-Extra: opencv
 Requires-Dist: opencv-python-headless>=4.8.0; extra == 'opencv'
+Provides-Extra: opencv-fips
+Requires-Dist: opencv-python-headless<4.13.0,>=4.8.0; extra == 'opencv-fips'
 Description-Content-Type: text/markdown
 
 <!-- markdownlint-disable MD041 -->
@@ -118,8 +124,8 @@ The guidance below applies to both the container and the Python-library forms.
 
 DataEval Flow is developed and tested on Linux (Ubuntu 22.04 and 24.04, including
 WSL2). The Python package supports **Python 3.10–3.14**; the CI test matrix runs
-all five. The container images are built on Ubuntu 22.04 (cu118) and
-Ubuntu 24.04 (cpu, cu128). macOS and Windows are supported only through Docker or
+all five. The container images are all built on Ubuntu 24.04 (cpu, cu126, cu130).
+macOS and Windows are supported only through Docker or
 WSL2 and are not part of the CI test matrix — if you hit an issue on those hosts,
 the OS/hardware may be the root cause.
 
@@ -158,17 +164,17 @@ at least the minimum CPU/memory above; size memory to your largest dataset.
 ## Quick Start
 
 ```bash
-# 1. Build CUDA 11.8 container
-docker build -f docker/Dockerfile.cu118 -t dataeval:cu118 .
+# 1. Build CUDA 12.6 container
+docker build -f docker/Dockerfile.cu126 -t dataeval:cu126 .
 
 # 2. Show help
-docker run dataeval:cu118
+docker run dataeval:cu126
 
 # 3. Run with data and output
 docker run --gpus all \
   --mount type=bind,source=/path/to/data,target=/dataeval,readonly \
   --mount type=bind,source=/path/to/output,target=/output \
-  dataeval:cu118
+  dataeval:cu126
 ```
 
 ## Pulling pre-built images
@@ -180,13 +186,13 @@ if you don't need to modify the code.
 **Rolling channel** — tracks the latest commit on `main`. The tag is overwritten on every merge.
 
 ```bash
-docker pull harbor.jatic.net/aria/dataeval:cu118   # cpu / cu118 / cu128
+docker pull harbor.jatic.net/aria/dataeval:cu126   # cpu / cu126 / cu130
 ```
 
 **Pinned release channel** — immutable, version-tagged images cut from `v*` git tags. Use these for reproducible workloads.
 
 ```bash
-docker pull harbor.jatic.net/aria/dataeval:0.1.0-cu118
+docker pull harbor.jatic.net/aria/dataeval:0.1.0-cu126
 ```
 
 **Verifying the signature** — every published image is signed with
@@ -194,11 +200,11 @@ docker pull harbor.jatic.net/aria/dataeval:0.1.0-cu118
 [docker/cosign.pub](docker/cosign.pub).
 
 ```bash
-cosign verify --key docker/cosign.pub harbor.jatic.net/aria/dataeval:cu118
+cosign verify --key docker/cosign.pub harbor.jatic.net/aria/dataeval:cu126
 ```
 
-Then drop the `dataeval:cu118` reference in the Quick Start `docker run`
-commands above with the fully-qualified `harbor.jatic.net/aria/dataeval:cu118`
+Then drop the `dataeval:cu126` reference in the Quick Start `docker run`
+commands above with the fully-qualified `harbor.jatic.net/aria/dataeval:cu126`
 (or pinned version) and skip step 1.
 
 > **Note on feature branches.** Containers are only built and published from
@@ -245,7 +251,7 @@ docker run --gpus all \
   --user "$(id -u):$(id -g)" \
   --mount type=bind,source=/path/to/data,target=/dataeval,readonly \
   --mount type=bind,source=/path/to/output,target=/output \
-  dataeval:cu118
+  dataeval:cu126
 ```
 
 #### Option 2: Open directory permissions
@@ -267,7 +273,7 @@ docker run --gpus all \
   -e DATAEVAL_DATA=/data \
   --mount type=bind,source=/path/to/data,target=/data,readonly \
   --mount type=bind,source=/path/to/output,target=/output \
-  dataeval:cu118
+  dataeval:cu126
 ```
 
 ## Environment Variables
@@ -305,7 +311,7 @@ options, precedence, and examples — via its help command, which is also the
 default when the container runs with no pipeline arguments:
 
 ```bash
-docker run dataeval:cu118 --help
+docker run dataeval:cu126 --help
 ```
 
 The library form exposes the same options via `python -m dataeval_flow --help`.
@@ -325,13 +331,13 @@ To specify a config path explicitly:
 docker run --gpus all \
   --mount type=bind,source=/path/to/data,target=/dataeval,readonly \
   --mount type=bind,source=/path/to/output,target=/output \
-  dataeval:cu118 --config config/
+  dataeval:cu126 --config config/
 
 # Single config file
 docker run --gpus all \
   --mount type=bind,source=/path/to/data,target=/dataeval,readonly \
   --mount type=bind,source=/path/to/output,target=/output \
-  dataeval:cu118 --config params.yaml
+  dataeval:cu126 --config params.yaml
 ```
 
 Dataset and model paths in config files are resolved relative to the data root (`/dataeval` by default).
@@ -444,8 +450,8 @@ DataEval Flow — it accepts the build already present (omit step 1 and you'll g
 CUDA-bundled manylinux build of torch from PyPI, which is much larger):
 
 ```bash
-# 1. Pick your PyTorch build (cpu / cu118 / cu128)
-pip install torch torchvision --index-url https://download.pytorch.org/whl/cu128
+# 1. Pick your PyTorch build (cpu / cu126 / cu130)
+pip install torch torchvision --index-url https://download.pytorch.org/whl/cu130
 
 # 2. Install DataEval Flow
 pip install dataeval-flow
@@ -463,14 +469,14 @@ pip install dataeval-flow
 `torchvision` is not installed by default — it is imported lazily and is only needed
 for preprocessing pipelines, the torchvision dataset adapter, and the TUI's transform
 discovery. Install it together with `torch` in step 1 so both come from the same index.
-Feature extras (`onnx`, `onnx-gpu`, `opencv`, `app`, `ontology`) work normally under
+Feature extras (`onnx`, `onnx-cu126`, `onnx-cu130`, `opencv`, `app`, `ontology`) work normally under
 pip and are independent of the PyTorch variant:
 
 ```bash
 pip install "dataeval-flow[onnx,opencv,app]"
 ```
 
-> **The `cpu` / `cu118` / `cu128` extras do not select a PyTorch variant under pip.**
+> **The `cpu` / `cu126` / `cu130` extras do not select a PyTorch variant under pip.**
 > All three declare the same requirements (`torch`, `torchvision`); what distinguishes
 > them is `[tool.uv.sources]`, which routes those packages to the right wheel index.
 > That is project metadata applied by uv when resolving **from source** — it is not
@@ -481,7 +487,7 @@ pip install "dataeval-flow[onnx,opencv,app]"
 `uv` from PyPI:
 
 ```bash
-uv pip install dataeval-flow --torch-backend cpu     # or cu118 / cu128 / auto
+uv pip install dataeval-flow --torch-backend cpu     # or cu126 / cu130 / auto
 ```
 
 `uv` from source (default toolchain; uses committed `uv.lock`) — extras apply here:
@@ -489,7 +495,7 @@ uv pip install dataeval-flow --torch-backend cpu     # or cu118 / cu128 / auto
 ```bash
 git clone https://github.com/aria-ml/dataeval-flow.git
 cd dataeval-flow
-uv sync --extra cpu      # or cu118 / cu128; add --extra onnx --extra opencv --extra app as needed
+uv sync --extra cpu      # or cu126 / cu130; add --extra onnx --extra opencv --extra app as needed
 ```
 
 `poetry` (source checkout; uses committed `poetry.lock`):
@@ -514,7 +520,7 @@ Notes:
 
 - PyTorch is installed from PyPI/`download.pytorch.org` in every path
   (it is no longer maintained on conda-forge).
-- GPU variants (`cu118`, `cu128`) are only wired through `uv` and
+- GPU variants (`cu126`, `cu130`) are only wired through `uv` and
   `pip` today; the Poetry/conda paths install the CPU build of PyTorch.
 
 **CLI Usage:**

{dataeval_flow-0.2.0 → dataeval_flow-0.2.1}/README.md

@@ -66,8 +66,8 @@ The guidance below applies to both the container and the Python-library forms.
 
 DataEval Flow is developed and tested on Linux (Ubuntu 22.04 and 24.04, including
 WSL2). The Python package supports **Python 3.10–3.14**; the CI test matrix runs
-all five. The container images are built on Ubuntu 22.04 (cu118) and
-Ubuntu 24.04 (cpu, cu128). macOS and Windows are supported only through Docker or
+all five. The container images are all built on Ubuntu 24.04 (cpu, cu126, cu130).
+macOS and Windows are supported only through Docker or
 WSL2 and are not part of the CI test matrix — if you hit an issue on those hosts,
 the OS/hardware may be the root cause.
 
@@ -106,17 +106,17 @@ at least the minimum CPU/memory above; size memory to your largest dataset.
 ## Quick Start
 
 ```bash
-# 1. Build CUDA 11.8 container
-docker build -f docker/Dockerfile.cu118 -t dataeval:cu118 .
+# 1. Build CUDA 12.6 container
+docker build -f docker/Dockerfile.cu126 -t dataeval:cu126 .
 
 # 2. Show help
-docker run dataeval:cu118
+docker run dataeval:cu126
 
 # 3. Run with data and output
 docker run --gpus all \
   --mount type=bind,source=/path/to/data,target=/dataeval,readonly \
   --mount type=bind,source=/path/to/output,target=/output \
-  dataeval:cu118
+  dataeval:cu126
 ```
 
 ## Pulling pre-built images
@@ -128,13 +128,13 @@ if you don't need to modify the code.
 **Rolling channel** — tracks the latest commit on `main`. The tag is overwritten on every merge.
 
 ```bash
-docker pull harbor.jatic.net/aria/dataeval:cu118   # cpu / cu118 / cu128
+docker pull harbor.jatic.net/aria/dataeval:cu126   # cpu / cu126 / cu130
 ```
 
 **Pinned release channel** — immutable, version-tagged images cut from `v*` git tags. Use these for reproducible workloads.
 
 ```bash
-docker pull harbor.jatic.net/aria/dataeval:0.1.0-cu118
+docker pull harbor.jatic.net/aria/dataeval:0.1.0-cu126
 ```
 
 **Verifying the signature** — every published image is signed with
@@ -142,11 +142,11 @@ docker pull harbor.jatic.net/aria/dataeval:0.1.0-cu118
 [docker/cosign.pub](docker/cosign.pub).
 
 ```bash
-cosign verify --key docker/cosign.pub harbor.jatic.net/aria/dataeval:cu118
+cosign verify --key docker/cosign.pub harbor.jatic.net/aria/dataeval:cu126
 ```
 
-Then drop the `dataeval:cu118` reference in the Quick Start `docker run`
-commands above with the fully-qualified `harbor.jatic.net/aria/dataeval:cu118`
+Then drop the `dataeval:cu126` reference in the Quick Start `docker run`
+commands above with the fully-qualified `harbor.jatic.net/aria/dataeval:cu126`
 (or pinned version) and skip step 1.
 
 > **Note on feature branches.** Containers are only built and published from
@@ -193,7 +193,7 @@ docker run --gpus all \
   --user "$(id -u):$(id -g)" \
   --mount type=bind,source=/path/to/data,target=/dataeval,readonly \
   --mount type=bind,source=/path/to/output,target=/output \
-  dataeval:cu118
+  dataeval:cu126
 ```
 
 #### Option 2: Open directory permissions
@@ -215,7 +215,7 @@ docker run --gpus all \
   -e DATAEVAL_DATA=/data \
   --mount type=bind,source=/path/to/data,target=/data,readonly \
   --mount type=bind,source=/path/to/output,target=/output \
-  dataeval:cu118
+  dataeval:cu126
 ```
 
 ## Environment Variables
@@ -253,7 +253,7 @@ options, precedence, and examples — via its help command, which is also the
 default when the container runs with no pipeline arguments:
 
 ```bash
-docker run dataeval:cu118 --help
+docker run dataeval:cu126 --help
 ```
 
 The library form exposes the same options via `python -m dataeval_flow --help`.
@@ -273,13 +273,13 @@ To specify a config path explicitly:
 docker run --gpus all \
   --mount type=bind,source=/path/to/data,target=/dataeval,readonly \
   --mount type=bind,source=/path/to/output,target=/output \
-  dataeval:cu118 --config config/
+  dataeval:cu126 --config config/
 
 # Single config file
 docker run --gpus all \
   --mount type=bind,source=/path/to/data,target=/dataeval,readonly \
   --mount type=bind,source=/path/to/output,target=/output \
-  dataeval:cu118 --config params.yaml
+  dataeval:cu126 --config params.yaml
 ```
 
 Dataset and model paths in config files are resolved relative to the data root (`/dataeval` by default).
@@ -392,8 +392,8 @@ DataEval Flow — it accepts the build already present (omit step 1 and you'll g
 CUDA-bundled manylinux build of torch from PyPI, which is much larger):
 
 ```bash
-# 1. Pick your PyTorch build (cpu / cu118 / cu128)
-pip install torch torchvision --index-url https://download.pytorch.org/whl/cu128
+# 1. Pick your PyTorch build (cpu / cu126 / cu130)
+pip install torch torchvision --index-url https://download.pytorch.org/whl/cu130
 
 # 2. Install DataEval Flow
 pip install dataeval-flow
@@ -411,14 +411,14 @@ pip install dataeval-flow
 `torchvision` is not installed by default — it is imported lazily and is only needed
 for preprocessing pipelines, the torchvision dataset adapter, and the TUI's transform
 discovery. Install it together with `torch` in step 1 so both come from the same index.
-Feature extras (`onnx`, `onnx-gpu`, `opencv`, `app`, `ontology`) work normally under
+Feature extras (`onnx`, `onnx-cu126`, `onnx-cu130`, `opencv`, `app`, `ontology`) work normally under
 pip and are independent of the PyTorch variant:
 
 ```bash
 pip install "dataeval-flow[onnx,opencv,app]"
 ```
 
-> **The `cpu` / `cu118` / `cu128` extras do not select a PyTorch variant under pip.**
+> **The `cpu` / `cu126` / `cu130` extras do not select a PyTorch variant under pip.**
 > All three declare the same requirements (`torch`, `torchvision`); what distinguishes
 > them is `[tool.uv.sources]`, which routes those packages to the right wheel index.
 > That is project metadata applied by uv when resolving **from source** — it is not
@@ -429,7 +429,7 @@ pip install "dataeval-flow[onnx,opencv,app]"
 `uv` from PyPI:
 
 ```bash
-uv pip install dataeval-flow --torch-backend cpu     # or cu118 / cu128 / auto
+uv pip install dataeval-flow --torch-backend cpu     # or cu126 / cu130 / auto
 ```
 
 `uv` from source (default toolchain; uses committed `uv.lock`) — extras apply here:
@@ -437,7 +437,7 @@ uv pip install dataeval-flow --torch-backend cpu     # or cu118 / cu128 / auto
 ```bash
 git clone https://github.com/aria-ml/dataeval-flow.git
 cd dataeval-flow
-uv sync --extra cpu      # or cu118 / cu128; add --extra onnx --extra opencv --extra app as needed
+uv sync --extra cpu      # or cu126 / cu130; add --extra onnx --extra opencv --extra app as needed
 ```
 
 `poetry` (source checkout; uses committed `poetry.lock`):
@@ -462,7 +462,7 @@ Notes:
 
 - PyTorch is installed from PyPI/`download.pytorch.org` in every path
   (it is no longer maintained on conda-forge).
-- GPU variants (`cu118`, `cu128`) are only wired through `uv` and
+- GPU variants (`cu126`, `cu130`) are only wired through `uv` and
   `pip` today; the Poetry/conda paths install the CPU build of PyTorch.
 
 **CLI Usage:**

{dataeval_flow-0.2.0 → dataeval_flow-0.2.1}/pyproject.toml

@@ -28,21 +28,31 @@ dependencies = [
 
 [project.optional-dependencies]
 cpu = ["torch>=2.2.0", "torchvision>=0.17.0"]
-cu118 = ["torch>=2.2.0", "torchvision>=0.17.0"]
-cu128 = ["torch>=2.2.0", "torchvision>=0.17.0"]
+cu126 = ["torch>=2.2.0", "torchvision>=0.17.0"]
+cu130 = ["torch>=2.2.0", "torchvision>=0.17.0"]
 onnx = [
   "onnx>=1.15",
   "onnxruntime>=1.20,<1.24; python_version == '3.10'",  # 1.24+ drops cp310 wheels
   "onnxruntime>=1.20; python_version >= '3.11'",
 ]
-onnx-gpu = [
+# One onnx extra per CUDA variant: an onnxruntime-gpu wheel links against a specific
+# CUDA major, so the runtime has to match the torch build (see docker/variants.yaml).
+# The `extra != ...` guards drop the requirement entirely if both are requested, which
+# keeps the published metadata resolvable under pip; uv enforces the exclusion outright
+# via [tool.uv].conflicts.
+onnx-cu126 = [
   "onnx>=1.15",
-  # 1.23.2 is the floor for the CUDA 12.8 images (see docker/variants.yaml).
-  "onnxruntime-gpu>=1.23.2,<1.24; python_version == '3.10'",  # 1.24+ drops cp310 wheels
-  "onnxruntime-gpu>=1.23.2,<1.27; python_version >= '3.11'",  # 1.27+ requires CUDA 13.0
-  "onnxruntime-gpu>=1.24,<1.27; python_version >= '3.14'",  # 1.24+ for cp314 wheels
+  "onnxruntime-gpu>=1.20,<1.24; python_version == '3.10' and extra != 'onnx-cu130'",  # 1.24+ drops cp310 wheels
+  "onnxruntime-gpu>=1.20,<1.27; python_version >= '3.11' and extra != 'onnx-cu130'",  # 1.27+ requires CUDA 13.0
+  "onnxruntime-gpu>=1.24,<1.27; python_version >= '3.14' and extra != 'onnx-cu130'",  # 1.24+ for cp314 wheels
 ]
-opencv = ["opencv-python-headless>=4.8.0"]
+onnx-cu130 = [
+  "onnx>=1.15",
+  "onnxruntime>=1.20; python_version == '3.10' and extra != 'onnx-cu126'",  # CUDA 13.0 is not supported on Python 3.10
+  "onnxruntime-gpu>=1.27; python_version >= '3.11' and extra != 'onnx-cu126'",
+]
+opencv = ["opencv-python-headless>=4.8.0"]  
+opencv-fips = ["opencv-python-headless>=4.8.0,<4.13.0"]  # OpenCV 4.13 and higher are not FIPS compliant
 app = ["textual>=3.0"]
 ontology = ["dataeval[ontology]==1.1.0-rc6"]
 
@@ -150,12 +160,17 @@ constraint-dependencies = [
 conflicts = [
   [
     { extra = "cpu" },
-    { extra = "cu118" },
-    { extra = "cu128" },
+    { extra = "cu126" },
+    { extra = "cu130" },
   ],
   [
     { extra = "onnx" },
-    { extra = "onnx-gpu" },
+    { extra = "onnx-cu126" },
+    { extra = "onnx-cu130" },
+  ],
+  [
+    { extra = "opencv" },
+    { extra = "opencv-fips" },
   ]
 ]
 
@@ -165,25 +180,25 @@ url = "https://download.pytorch.org/whl/cpu"
 explicit = true
 
 [[tool.uv.index]]
-name = "pytorch-cu118"
-url = "https://download.pytorch.org/whl/cu118"
+name = "pytorch-cu126"
+url = "https://download.pytorch.org/whl/cu126"
 explicit = true
 
 [[tool.uv.index]]
-name = "pytorch-cu128"
-url = "https://download.pytorch.org/whl/cu128"
+name = "pytorch-cu130"
+url = "https://download.pytorch.org/whl/cu130"
 explicit = true
 
 [tool.uv.sources]
 torch = [
   { index = "pytorch-cpu", extra = "cpu" },
-  { index = "pytorch-cu118", extra = "cu118" },
-  { index = "pytorch-cu128", extra = "cu128" },
+  { index = "pytorch-cu126", extra = "cu126" },
+  { index = "pytorch-cu130", extra = "cu130" },
 ]
 torchvision = [
   { index = "pytorch-cpu", extra = "cpu" },
-  { index = "pytorch-cu118", extra = "cu118" },
-  { index = "pytorch-cu128", extra = "cu128" },
+  { index = "pytorch-cu126", extra = "cu126" },
+  { index = "pytorch-cu130", extra = "cu130" },
 ]
 
 [tool.poetry]

{dataeval_flow-0.2.0 → dataeval_flow-0.2.1}/src/dataeval_flow/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '0.2.0'
-__version_tuple__ = version_tuple = (0, 2, 0)
+__version__ = version = '0.2.1'
+__version_tuple__ = version_tuple = (0, 2, 1)
 
 __commit_id__ = commit_id = None

{dataeval_flow-0.2.0 → dataeval_flow-0.2.1}/src/dataeval_flow/config/schemas/_dataset.py

@@ -79,15 +79,25 @@ class CocoDatasetConfig(_DatasetConfigBase):
 class YoloDatasetConfig(_DatasetConfigBase):
     """Dataset config for YOLO format.
 
+    ``path`` is the dataset root — the directory holding ``data.yaml`` and the
+    image/label trees — for either Ultralytics arrangement (``images/train/`` +
+    ``labels/train/`` or ``train/images/`` + ``train/labels/``).  Select a split
+    with ``split`` rather than by pointing ``path`` at a split subdirectory,
+    which puts ``data.yaml`` out of scope and falls back to numeric class names.
+
     YAML example::
 
         datasets:
           - name: yolo_train
             format: yolo
             path: yolo
+            split: train
     """
 
     format: Literal["yolo"] = "yolo"
+    split: str | None = None
+    yaml_file: str | None = None
+    ann_dir: str | None = None
 
 
 class DatasetProtocolConfig(BaseModel):

{dataeval_flow-0.2.0 → dataeval_flow-0.2.1}/src/dataeval_flow/dataset.py

@@ -408,15 +408,54 @@ def load_dataset_coco(path: Path, *, annotations_file: str | None = None, images
     return dataset
 
 
-def load_dataset_yolo(path: Path) -> Any:
+def load_dataset_yolo(
+    path: Path,
+    *,
+    split: str | None = None,
+    yaml_file: str | None = None,
+    ann_dir: str | None = None,
+) -> Any:
     """Load a YOLO-format object detection dataset via datamaite.
 
-    datamaite expects the standard ``images/`` + ``labels/`` + ``data.yaml`` layout.
+    datamaite expects the standard ``images/`` + ``labels/`` + ``data.yaml``
+    layout, in either Ultralytics arrangement (``images/train/`` +
+    ``labels/train/`` or ``train/images/`` + ``train/labels/``), as well as the
+    split-less ``images/`` + ``labels/`` variant.
+
+    Parameters
+    ----------
+    path : Path
+        Dataset root — the directory holding ``data.yaml`` and the image/label
+        trees.  Always point at the root rather than a split subdirectory: the
+        root is where the class names are found.
+    split : str | None
+        Load only this split (``"train"``/``"val"``/``"test"``; aliases such as
+        ``"validation"`` normalize).  Defaults to every split under *path*.
+    yaml_file : str | None
+        ``data.yaml`` path relative to *path*, for a config that is not at the
+        root under a conventional name.  It is authoritative — a missing file
+        or one whose image sources yield nothing gives an empty dataset rather
+        than falling back to a scan of *path*.
+    ann_dir : str | None
+        Label directory relative to *path*, for label trees kept outside the
+        conventional ``labels/`` sibling of ``images/``.
     """
     from datamaite import load_od
 
-    dataset = load_od(path, dataset_format="yolo")
-    _logger.info("YOLO dataset: loaded %d images from %s", len(dataset), path)
+    kwargs: dict[str, str] = {}
+    if split is not None:
+        kwargs["split"] = split
+    if yaml_file is not None:
+        kwargs["yaml_file"] = yaml_file
+    if ann_dir is not None:
+        kwargs["ann_dir"] = ann_dir
+    dataset = load_od(path, dataset_format="yolo", **kwargs)
+    _logger.info(
+        "YOLO dataset: loaded %d images from %s%s",
+        len(dataset),
+        path,
+        f" (split={split!r})" if split is not None else "",
+    )
     return dataset
 
 
@@ -430,6 +469,8 @@ def load_dataset(
     task: Literal["image_classification", "object_detection"] = "image_classification",
     annotations_file: str | None = None,
     images_dir: str | None = None,
+    yaml_file: str | None = None,
+    ann_dir: str | None = None,
 ) -> Any:
     """Load a dataset and convert to MAITE format.
 
@@ -441,7 +482,10 @@ def load_dataset(
     path : Path
         Path to the dataset directory.
     split : str | None
-        Optional split name to load (e.g. "train", "test").
+        Optional split name to load (e.g. "train", "val").  HuggingFace treats
+        it as a subdirectory of *path*; YOLO selects among the splits found
+        under the dataset root.  Unused by the other formats — COCO selects a
+        split through ``annotations_file``.
     dataset_format : Literal["huggingface", "coco", "yolo", "image_folder"]
         Dataset format identifier (default ``"huggingface"``).
     recursive : bool
@@ -454,6 +498,10 @@ def load_dataset(
         Annotations file name (COCO only).
     images_dir : str | None
         Images subdirectory name (COCO only).
+    yaml_file : str | None
+        ``data.yaml`` path relative to *path* (YOLO only).
+    ann_dir : str | None
+        Label directory relative to *path* (YOLO only).
 
     Returns
     -------
@@ -478,16 +526,19 @@ def load_dataset(
     elif dataset_format == "coco":
         dataset = load_dataset_coco(path, annotations_file=annotations_file, images_dir=images_dir)
     elif dataset_format == "yolo":
-        dataset = load_dataset_yolo(path)
+        dataset = load_dataset_yolo(path, split=split, yaml_file=yaml_file, ann_dir=ann_dir)
     else:
         msg = f"Unsupported dataset format: {dataset_format!r}"
         raise ValueError(msg)
 
-    _reject_empty_dataset(dataset, path / split if split else path, dataset_format)
+    # Only HuggingFace resolves a split to a subdirectory; for YOLO the split is
+    # selected inside the root, so the root is what an error should name.
+    root = path / split if dataset_format == "huggingface" and split else path
+    _reject_empty_dataset(dataset, root, dataset_format, split=split)
     return dataset
 
 
-def _reject_empty_dataset(dataset: Any, root: Path, dataset_format: str) -> None:
+def _reject_empty_dataset(dataset: Any, root: Path, dataset_format: str, split: str | None = None) -> None:
     """Raise when a loader yielded zero items.
 
     datamaite's loaders log a warning and return an empty dataset when the
@@ -503,8 +554,16 @@ def _reject_empty_dataset(dataset: Any, root: Path, dataset_format: str) -> None
     if len(dataset) > 0:
         return
 
-    msg = f"Loaded 0 items from {root} (format={dataset_format!r})."
-    if dataset_format == "huggingface" and _is_arrow_dump(root):
+    detail = f"format={dataset_format!r}" + (f", split={split!r}" if split else "")
+    msg = f"Loaded 0 items from {root} ({detail})."
+    if dataset_format == "yolo" and split:
+        msg += (
+            f" No images matched split {split!r} — a YOLO split selection that matches"
+            " nothing selects nothing rather than widening back to every split. Check"
+            f" that the root holds an images/{split}/ or {split}/images/ tree, or a"
+            " data.yaml declaring that split."
+        )
+    elif dataset_format == "huggingface" and _is_arrow_dump(root):
         msg += (
             " The directory is a `datasets.save_to_disk()` Arrow dump, which the"
             " huggingface loader does not read — it expects the local ImageFolder"
@@ -569,9 +628,11 @@ def resolve_dataset(config: BaseModel, data_dir: Path | None = None) -> Resolved
     """
     from dataeval_flow.cache import dataset_fingerprint
     from dataeval_flow.config.schemas._dataset import (
+        CocoDatasetConfig,
         DatasetProtocolConfig,
         HuggingFaceDatasetConfig,
         ImageFolderDatasetConfig,
+        YoloDatasetConfig,
         _DatasetConfigBase,
     )
 
@@ -588,11 +649,13 @@ def resolve_dataset(config: BaseModel, data_dir: Path | None = None) -> Resolved
         elif isinstance(config, ImageFolderDatasetConfig):
             kwargs["recursive"] = config.recursive
             kwargs["infer_labels"] = config.infer_labels
-        else:
-            # Coco forwards annotations_file/images_dir; Yolo forwards nothing.
-            for field_name in ("annotations_file", "images_dir"):
-                if hasattr(config, field_name):
-                    kwargs[field_name] = getattr(config, field_name)
+        elif isinstance(config, CocoDatasetConfig):
+            kwargs["annotations_file"] = config.annotations_file
+            kwargs["images_dir"] = config.images_dir
+        elif isinstance(config, YoloDatasetConfig):
+            kwargs["split"] = config.split
+            kwargs["yaml_file"] = config.yaml_file
+            kwargs["ann_dir"] = config.ann_dir
 
         from dataeval_flow.config._loader import resolve_path