Updates to various components, tests, and configs to pass tests within the slim Docker image, all of which appear to be real errors.

Signed-off-by: Eric Kerfoot <[email protected]>

Author: ericspod

Dockerfile.slim

@@ -14,25 +14,21 @@
 ARG IMAGE=debian:12-slim
 FROM ${IMAGE} AS build
 
-LABEL maintainer="[email protected]"
+ENV DEBIAN_FRONTEND=noninteractive
+ENV APT_INSTALL="apt install -y --no-install-recommends"
 
-RUN export DEBIAN_FRONTEND=noninteractive && \
-    apt update && \
-    apt upgrade -y && \
-    apt install -y --no-install-recommends ca-certificates python3-pip python-is-python3 git wget libopenslide0 unzip python3-dev && \
+RUN apt update && apt upgrade -y && \
+    ${APT_INSTALL} ca-certificates python3-pip python-is-python3 git wget libopenslide0 unzip python3-dev && \
     wget https://developer.download.nvidia.com/compute/cuda/repos/debian12/x86_64/cuda-keyring_1.1-1_all.deb && \
     dpkg -i cuda-keyring_1.1-1_all.deb && \
     apt update && \
-    apt install -y --no-install-recommends cuda-toolkit-12 && \
+    ${APT_INSTALL} cuda-toolkit-12 && \
     apt clean && \
     rm -rf cuda-keyring_1.1-1_all.deb /usr/lib/python*/EXTERNALLY-MANAGED /var/lib/apt/lists/* 
 
 # TODO: remark for issue [revise the dockerfile](https://github.com/zarr-developers/numcodecs/issues/431)
 RUN if [[ $(uname -m) =~ "aarch64" ]]; then \
-      export CFLAGS="-O3" && \
-      export DISABLE_NUMCODECS_SSE2=true && \
-      export DISABLE_NUMCODECS_AVX2=true && \
-      python -m pip install numcodecs; \
+      CFLAGS="-O3" DISABLE_NUMCODECS_SSE2=true DISABLE_NUMCODECS_AVX2=true python -m pip install numcodecs; \
     fi
 
 # NGC Client
@@ -41,8 +37,6 @@ ARG NGC_CLI_URI="https://ngc.nvidia.com/downloads/ngccli_linux.zip"
 RUN wget -q ${NGC_CLI_URI} && unzip ngccli_linux.zip && chmod u+x ngc-cli/ngc && \
     find ngc-cli/ -type f -exec md5sum {} + | LC_ALL=C sort | md5sum -c ngc-cli.md5 && \
     rm -rf ngccli_linux.zip ngc-cli.md5
-ENV PATH=${PATH}:/opt/tools:/opt/tools/ngc-cli
-ENV POLYGRAPHY_AUTOINSTALL_DEPS=1
 
 WORKDIR /opt/monai
 
@@ -56,12 +50,38 @@ COPY monai ./monai
 RUN python -m pip install --upgrade --no-cache-dir pip && \
     python -m pip install --no-cache-dir -r requirements-dev.txt 
 
+RUN wget https://developer.download.nvidia.com/compute/cuda/repos/debian12/x86_64/cuda-keyring_1.1-1_all.deb && \
+    dpkg -i cuda-keyring_1.1-1_all.deb && \
+    apt update && \
+    ${APT_INSTALL} cuda-toolkit-12
+
 # compile ext and remove temp files
 RUN export TORCH_CUDA_ARCH_LIST=Turing CUDA_HOME=/usr/local/cuda BUILD_MONAI=1 FORCE_CUDA=1 USE_COMPILED=1 && \
     python setup.py develop && \
-    rm -rf build __pycache__ monai.egg-info
+    rm -rf build __pycache__ monai.egg-info 
 
 # flatten all layers down to one
 FROM ${IMAGE}
+LABEL maintainer="[email protected]"
+
 WORKDIR /opt/monai
 COPY --from=build / /
+
+ENV PATH=${PATH}:/opt/tools:/opt/tools/ngc-cli
+ENV POLYGRAPHY_AUTOINSTALL_DEPS=1
+ENV CUDA_HOME=/usr/local/cuda
+ENV BUILD_MONAI=1
+
+# docker run -ti --rm --gpus '"device=0,1"' --shm-size=10gb monai_slim /bin/bash
+
+# PYTHONPATH=. BUILD_MONAI=1 ./runtests.sh -u 2>&1|tee tests.log
+# PYTHONPATH=. BUILD_MONAI=1 python -m unittest tests.networks.layers.filtering.test_trainable_bilateral
+# PYTHONPATH=. BUILD_MONAI=1 python -m unittest tests.apps.detection.networks.test_retinanet.TestRetinaNet.test_onnx_0
+# PYTHONPATH=. BUILD_MONAI=1 python -m unittest tests.integration.test_auto3dseg_hpo.TestHPO.test_run_algo
+# PYTHONPATH=. BUILD_MONAI=1 python -m unittest tests.integration.test_auto3dseg_hpo.TestHPO.test_get_history
+# PYTHONPATH=. BUILD_MONAI=1 python -m unittest tests.networks.layers.test_gmm
+# python -m unittest tests.apps.vista3d.test_point_based_window_inferer
+# python -m unittest tests.bundle.test_bundle_download.TestDownload.test_ngc_private_source_download_bundle_0
+# python -m unittest tests.data.meta_tensor.test_meta_tensor.TestMetaTensor.test_pickling
+# PYTHONPATH=. BUILD_MONAI=1 python -m unittest tests.networks.test_convert_to_onnx
+

monai/apps/vista3d/inferer.py

@@ -86,13 +86,13 @@ def point_based_window_inferer(
             for j in range(len(ly_)):
                 for k in range(len(lz_)):
                     lx, rx, ly, ry, lz, rz = (lx_[i], rx_[i], ly_[j], ry_[j], lz_[k], rz_[k])
-                    unravel_slice = [
+                    unravel_slice = (
                         slice(None),
                         slice(None),
                         slice(int(lx), int(rx)),
                         slice(int(ly), int(ry)),
                         slice(int(lz), int(rz)),
-                    ]
+                    )
                     batch_image = image[unravel_slice]
                     output = predictor(
                         batch_image,

monai/bundle/scripts.py

@@ -515,7 +515,7 @@ def download(
             bundles for "monaihosting" source are also hosted on Hugging Face Hub, but the "repo_id" is always in the form
             of "MONAI/bundle_name", therefore, this argument is not required for "monaihosting" source.
             If `source` is "ngc_private", it should be in the form of "org/org_name" or "org/org_name/team/team_name",
-            or you can specify the environment variable NGC_ORG and NGC_TEAM.
+            or you can specify the environment variable NGC_ORG and NGC_TEAM. 
         url: url to download the data. If not `None`, data will be downloaded directly
             and `source` will not be checked.
             If `name` is `None`, filename is determined by `monai.apps.utils._basename(url)`.

monai/networks/nets/vista3d.py

@@ -243,14 +243,10 @@ def connected_components_combine(
         _logits = logits[mapping_index]
         inside = []
         for i in range(_logits.shape[0]):
-            inside.append(
-                np.any(
-                    [
-                        _logits[i, 0, p[0], p[1], p[2]].item() > 0
-                        for p in point_coords[i].cpu().numpy().round().astype(int)
-                    ]
-                )
-            )
+            p_coord = point_coords[i].cpu().numpy().round().astype(int)
+            inside_p = [_logits[i, 0, p[0], p[1], p[2]].item() > 0 for p in p_coord]
+            inside.append(int(np.any(inside_p)))  # convert to int to avoid typing problems with Numpy
+
         inside_tensor = torch.tensor(inside).to(logits.device)
         nan_mask = torch.isnan(_logits)
         # _logits are converted to binary [B1, 1, H, W, D]

monai/networks/utils.py

@@ -713,13 +713,13 @@ def convert_to_onnx(
         torch_versioned_kwargs = {}
         if use_trace:
             # let torch.onnx.export to trace the model.
-            mode_to_export = model
+            model_to_export = model
             torch_versioned_kwargs = kwargs
             if "dynamo" in kwargs and kwargs["dynamo"] and verify:
                 torch_versioned_kwargs["verify"] = verify
                 verify = False
         else:
-            mode_to_export = torch.jit.script(model, **kwargs)
+            model_to_export = torch.jit.script(model, **kwargs)
 
         if torch.is_tensor(inputs) or isinstance(inputs, dict):
             onnx_inputs = (inputs,)
@@ -733,7 +733,7 @@ def convert_to_onnx(
             f = filename
         print(f"torch_versioned_kwargs={torch_versioned_kwargs}")
         torch.onnx.export(
-            mode_to_export,
+            model_to_export,
             onnx_inputs,
             f=f,
             input_names=input_names,

requirements-dev.txt

@@ -1,6 +1,6 @@
 # Full requirements for developments
 -r requirements-min.txt
-pytorch-ignite==0.4.11
+pytorch-ignite
 gdown>=4.7.3
 scipy>=1.12.0; python_version >= '3.9'
 itk>=5.2
@@ -51,8 +51,9 @@ h5py
 nni==2.10.1; platform_system == "Linux" and "arm" not in platform_machine and "aarch" not in platform_machine
 optuna
 git+https://github.com/Project-MONAI/MetricsReloaded@monai-support#egg=MetricsReloaded
-onnx>=1.13.0, <1.19.1
-onnxruntime; python_version <= '3.10'
+onnx>=1.13.0, <1.19.0
+onnxruntime
+onnxscript
 typeguard<3  # https://github.com/microsoft/nni/issues/5457
 filelock<3.12.0  # https://github.com/microsoft/nni/issues/5523
 zarr

requirements.txt

@@ -1,3 +1,3 @@
-torch>=2.4.1; platform_system != "Windows"
-torch>=2.4.1, !=2.7.0; platform_system == "Windows"
+torch>=2.4.1, <2.9; platform_system != "Windows"
+torch>=2.4.1, <2.9, !=2.7.0; platform_system == "Windows"
 numpy>=1.24,<3.0

tests/bundle/test_bundle_download.py

@@ -15,7 +15,7 @@
 import os
 import tempfile
 import unittest
-from unittest.case import skipUnless
+from unittest.case import skipIf, skipUnless
 from unittest.mock import patch
 
 import numpy as np
@@ -219,6 +219,7 @@ def test_monaihosting_url_download_bundle(self, bundle_files, bundle_name, url):
 
     @parameterized.expand([TEST_CASE_5])
     @skip_if_quick
+    @skipIf(os.getenv("NGC_API_KEY", None) == None, "NGC API key required for this test")
     def test_ngc_private_source_download_bundle(self, bundle_files, bundle_name, _url):
         with skip_if_downloading_fails():
             # download a single file from url, also use `args_file`

tests/data/meta_tensor/test_meta_tensor.py

@@ -245,7 +245,7 @@ def test_pickling(self):
         with tempfile.TemporaryDirectory() as tmp_dir:
             fname = os.path.join(tmp_dir, "im.pt")
             torch.save(m, fname)
-            m2 = torch.load(fname, weights_only=True)
+            m2 = torch.load(fname, weights_only=False)
         self.check(m2, m, ids=False)
 
     @skip_if_no_cuda

tests/losses/test_multi_scale.py

@@ -55,7 +55,7 @@ class TestMultiScale(unittest.TestCase):
     @parameterized.expand(TEST_CASES)
     def test_shape(self, input_param, input_data, expected_val):
         result = MultiScaleLoss(**input_param).forward(**input_data)
-        np.testing.assert_allclose(result.detach().cpu().numpy(), expected_val, rtol=1e-5)
+        np.testing.assert_allclose(result.detach().cpu().numpy(), expected_val, rtol=1e-4)
 
     @parameterized.expand(
         [
@@ -84,3 +84,4 @@ def test_script(self):
 
 if __name__ == "__main__":
     unittest.main()
+