releasing minor 1.6.2 [rebase & merge]

.azure/gpu-integrations.yml

@@ -22,8 +22,8 @@ jobs:
           torch-ver: "2.0"
           requires: "oldest"
         "torch | 2.x":
-          docker-image: "pytorch/pytorch:2.5.0-cuda12.1-cudnn9-runtime"
-          torch-ver: "2.5"
+          docker-image: "pytorch/pytorch:2.6.0-cuda12.4-cudnn9-runtime"
+          torch-ver: "2.6"
     # how long to run the job before automatically cancelling
     timeoutInMinutes: "40"
     # how much time to give 'run always even if cancelled tasks' before stopping them

.azure/gpu-unittests.yml

@@ -29,9 +29,6 @@ jobs:
           docker-image: "ubuntu22.04-cuda11.8.0-py3.10-torch2.0"
           torch-ver: "2.0"
         "PyTorch | 2.X stable":
-          docker-image: "ubuntu22.04-cuda12.1.1-py3.11-torch2.5"
-          torch-ver: "2.5"
-        "PyTorch | 2.X future":
           docker-image: "ubuntu22.04-cuda12.4.1-py3.11-torch2.6"
           torch-ver: "2.6"
     # how long to run the job before automatically cancelling

.devcontainer/Dockerfile

@@ -19,7 +19,7 @@ RUN \
     pip3 --disable-pip-version-check --no-cache-dir install \
       -r /tmp/pip-tmp/requirements/_devel.txt \
       -r /tmp/pip-tmp/requirements/_docs.txt \
-      --find-links="https://download.pytorch.org/whl/cpu/torch_stable.html" \
+      --extra-index-url="https://download.pytorch.org/whl/cpu/" \
       --find-links="dist/" && \
     rm -rf /tmp/pip-tmp
 

.github/actions/pull-caches/action.yml

@@ -18,6 +18,14 @@ inputs:
     description: cache restore/dump key
     required: false
     default: "pypi-packages"
+  cache-torch-HF:
+    description: "cache torch and HF"
+    required: false
+    default: "true"
+  cache-references:
+    description: "cache metrics references"
+    required: false
+    default: "false"
 
 runs:
   using: "composite"
@@ -67,6 +75,7 @@ runs:
       shell: bash
 
     - name: Cache Torch & HF
+      if: inputs.cache-torch-HF == 'true' # since the input is string
       continue-on-error: true
       uses: actions/cache/restore@v3
       with:
@@ -75,6 +84,7 @@ runs:
         key: ci-caches
 
     - name: Restored Torch & HF
+      if: inputs.cache-torch-HF == 'true' # since the input is string
       run: |
         mkdir -p $CACHES_DIR
         pip install -q py-tree
@@ -83,14 +93,15 @@ runs:
 
     - name: Cache References
       # do not use this cache for dispatch and crone, to enable rebuild caches if needed
-      if: github.event_name != 'workflow_dispatch' && github.event_name != 'schedule'
+      if: github.event_name != 'workflow_dispatch' && github.event_name != 'schedule' && inputs.cache-references == 'true'
       continue-on-error: true
       uses: actions/cache/restore@v3
       with:
         path: tests/_cache-references
         key: cache-references
 
     - name: Restored References
+      if: inputs.cache-references == 'true' # since the input is string
       continue-on-error: true
       working-directory: tests/
       run: |

.github/actions/push-caches/action.yml

@@ -14,6 +14,18 @@ inputs:
     description: location to pull PyTorch from
     required: false
     default: "https://download.pytorch.org/whl/cpu/torch_stable.html"
+  cache-artifact-appendix:
+    description: "unique name or running index"
+    required: false
+    default: ""
+  cache-torch-HF:
+    description: "cache torch and HF"
+    required: false
+    default: "true"
+  cache-references:
+    description: "cache metrics references"
+    required: false
+    default: "false"
 
 runs:
   using: "composite"
@@ -23,66 +35,50 @@ runs:
       shell: bash
 
     - name: Freeze local emv.
+      if: inputs.cache-artifact-appendix != ''
       run: |
         pip freeze > requirements.dump
         cat requirements.dump
       shell: bash
 
-    #- name: Filter self pkg
-    #  run: |
-    #    import os
-    #    fp = 'requirements.dump'
-    #    with open(fp) as fopen:
-    #      lines = [ln.strip() for ln in fopen.readlines()]
-    #    lines = [ln.split('+')[0] for ln in lines if '-e ' not in ln]
-    #    with open(fp, 'w') as fwrite:
-    #      fwrite.writelines([ln + os.linesep for ln in lines])
-    #  shell: python
-
     - name: Dump wheels
+      if: inputs.cache-artifact-appendix != ''
       run: |
         pip wheel -r requirements/_devel.txt --prefer-binary \
-          --wheel-dir=.pip-wheels \
+          --wheel-dir=_pip-wheels \
           -f ${{ inputs.torch-url }} -f ${{ inputs.pypi-dir }}
-        ls -lh .pip-wheels
+        ls -lh _pip-wheels
       shell: bash
 
-    - name: Cache pull packages
-      uses: actions/cache/restore@v3
-      with:
-        enableCrossOsArchive: true
-        path: ${{ inputs.pypi-dir }}
-        key: ${{ inputs.pypi-key }}
-
-    - name: Find diff
-      id: wheels-diff
+    - name: Move new packages to staging
+      if: inputs.cache-artifact-appendix != ''
       run: |
-        import os, glob
-        wheels = [os.path.basename(p) for p in glob.glob(".pip-wheels/*")]
-        pkgs = [os.path.basename(p) for p in glob.glob("${{ inputs.pypi-dir }}/*")]
-        diff = [w for w in wheels if w not in pkgs]
-        print(diff)
-        with open(os.environ['GITHUB_OUTPUT'], 'a') as fh:
-            print(f'count-new={len(diff)}', file=fh)
-      shell: python
-
-    - run: cp .pip-wheels/* ${{ inputs.pypi-dir }}
-      if: ${{ steps.wheels-diff.outputs.count-new != 0 }}
+        mkdir -p _pip-staging
+        python .github/assistant.py move_new_packages \
+          --dir-cache="${{ inputs.pypi-dir }}" \
+          --dir_local="_pip-wheels" \
+          --dir_staging="_pip-staging"
+        ls -lh _pip-staging/
+        # count files in the staging dir
+        file_count=$(ls -1 "_pip-staging/" | wc -l)
+        echo "NUM_PACKAGES=$file_count" >> $GITHUB_ENV
       shell: bash
 
-    - name: Cache push packages
-      if: ${{ steps.wheels-diff.outputs.count-new != 0 }}
-      uses: actions/cache/save@v3
+    - name: Upload new packages
+      if: inputs.cache-artifact-appendix != '' && env.NUM_PACKAGES != 0
+      uses: actions/upload-artifact@v4
       with:
-        enableCrossOsArchive: true
-        path: ${{ inputs.pypi-dir }}
-        key: ${{ inputs.pypi-key }}
+        name: ${{ inputs.pypi-key }}-run-${{ inputs.cache-artifact-appendix }}
+        path: _pip-staging
+        retention-days: 1
 
     - name: Post Torch & HF
+      if: inputs.cache-torch-HF == 'true' # since the input is string
       run: py-tree $CACHES_DIR
       shell: bash
 
     - name: Cache Torch & HF
+      if: inputs.cache-torch-HF == 'true' # since the input is string
       continue-on-error: true
       uses: actions/cache/save@v3
       with:
@@ -91,13 +87,16 @@ runs:
         key: ci-caches
 
     - name: Cache references
+      if: inputs.cache-references == 'true' # since the input is string
       continue-on-error: true
       uses: actions/cache/save@v3
       with:
         #enableCrossOsArchive: true
         path: tests/_cache-references
         key: cache-references
 
-    - name: Post References
-      run: py-tree tests/_cache-references/ --show_hidden
-      shell: bash
+    #- name: Post References
+    #  # This print taken soo many lines, so it is commented out
+    #  if: inputs.cache-references == 'true' # since the input is string
+    #  run: py-tree tests/_cache-references/ --show_hidden
+    #  shell: bash

.github/assistant.py

@@ -20,7 +20,6 @@
 
 import fire
 from packaging.version import parse
-from pkg_resources import parse_requirements
 
 _REQUEST_TIMEOUT = 10
 _PATH_ROOT = os.path.dirname(os.path.dirname(__file__))
@@ -58,6 +57,9 @@ def set_min_torch_by_python(fpath: str = "requirements/base.txt") -> None:
         >>> AssistantCLI.set_min_torch_by_python("../requirements/base.txt")
 
         """
+        # ToDo: `pkg_resources` is deprecates and shall be updated
+        from pkg_resources import parse_requirements
+
         py_ver = f"{sys.version_info.major}.{sys.version_info.minor}"
         if py_ver not in LUT_PYTHON_TORCH:
             return
@@ -114,52 +116,102 @@ def changed_domains(
         """Determine what domains were changed in particular PR."""
         import github
 
+        # define some edge case return cases
+        _return_all = "unittests" if not as_list else ["torchmetrics"]
+        _return_empty = [] if as_list else ""
+
+        # early return if no PR number
         if not pr:
-            return "unittests"
-        gh = github.Github()
+            return _return_all
+        gh = github.Github(login_or_token=auth_token)
         pr = gh.get_repo("Lightning-AI/torchmetrics").get_pull(pr)
         files = [f.filename for f in pr.get_files()]
 
         # filter out all integrations as they run in separate suit
         files = [fn for fn in files if not fn.startswith("tests/integrations")]
         if not files:
             logging.debug("Only integrations was changed so not reason for deep testing...")
-            return ""
+            return _return_empty
+
         # filter only docs files
-        files_ = [fn for fn in files if fn.startswith("docs")]
-        if len(files) == len(files_):
+        files_docs = [fn for fn in files if fn.startswith("docs")]
+        if len(files) == len(files_docs):
             logging.debug("Only docs was changed so not reason for deep testing...")
-            return ""
+            return _return_empty
+
+        files_markdown = [fn for fn in files if fn.endswith(".md")]
+        if len(files) == len(files_markdown):
+            logging.debug("Only markdown files was changed so not reason for deep testing...")
+            return _return_empty
+        # filter out markdown files
+        files = [fn for fn in files if fn not in files_markdown]
+
+        # filter only testing files which are not specific tests so for example configurations or helper tools
+        files_testing = [fn for fn in files if fn.startswith("tests") and not fn.endswith(".md") and "test_" not in fn]
+        if files_testing:
+            logging.debug("Some testing files was changed -> rather test everything...")
+            return _return_all
+
+        # files in requirements folder
+        files_req = [fn for fn in files if fn.startswith("requirements")]
+        req_domains = [fn.split("/")[1] for fn in files_req]
+        # cleaning up determining domains
+        req_domains = [req.replace(".txt", "").replace("_test", "") for req in req_domains if not req.endswith("_")]
+        # if you touch base, you need to run everything
+        if "base" in req_domains:
+            return _return_all
 
         # filter only package files and skip inits
-        _is_in_test = lambda fn: fn.startswith("tests")
-        _filter_pkg = lambda fn: _is_in_test(fn) or (fn.startswith("src/torchmetrics") and "__init__.py" not in fn)
+        _is_in_test = lambda fname: fname.startswith("tests")
+        _filter_pkg = lambda fname: _is_in_test(fname) or (
+            fname.startswith("src/torchmetrics") and "__init__.py" not in fname
+        )
         files_pkg = [fn for fn in files if _filter_pkg(fn)]
         if not files_pkg:
-            return "unittests"
+            return _return_all
 
         # parse domains
-        def _crop_path(fname: str, paths: list[str]) -> str:
+        def _crop_path(fname: str, paths: tuple[str] = ("src/torchmetrics/", "tests/unittests/", "functional/")) -> str:
             for p in paths:
                 fname = fname.replace(p, "")
             return fname
 
-        files_pkg = [_crop_path(fn, ["src/torchmetrics/", "tests/unittests/", "functional/"]) for fn in files_pkg]
+        files_pkg = [_crop_path(fn) for fn in files_pkg]
         # filter domain names
         tm_modules = [fn.split("/")[0] for fn in files_pkg if "/" in fn]
         # filter general (used everywhere) sub-packages
         tm_modules = [md for md in tm_modules if md not in general_sub_pkgs]
         if len(files_pkg) > len(tm_modules):
             logging.debug("Some more files was changed -> rather test everything...")
-            return "unittests"
-        # keep only unique
-        if as_list:
-            return list(tm_modules)
-        tm_modules = [f"unittests/{md}" for md in set(tm_modules)]
-        not_exists = [p for p in tm_modules if os.path.exists(p)]
+            return _return_all
+
+        # compose the final list with requirements and touched modules
+        test_modules = set(tm_modules + list(req_domains))
+        if as_list:  # keep only unique
+            return list(test_modules)
+
+        test_modules = [f"unittests/{md}" for md in set(test_modules)]
+        not_exists = [p for p in test_modules if os.path.exists(p)]
         if not_exists:
             raise ValueError(f"Missing following paths: {not_exists}")
-        return " ".join(tm_modules)
+        return " ".join(test_modules)
+
+    @staticmethod
+    def move_new_packages(dir_cache: str, dir_local: str, dir_staging: str) -> None:
+        """Move unique packages from local folder to staging."""
+        assert os.path.isdir(dir_cache), f"Missing folder with saved packages: '{dir_cache}'"  # noqa: S101
+        assert os.path.isdir(dir_local), f"Missing folder with local packages: '{dir_local}'"  # noqa: S101
+        assert os.path.isdir(dir_staging), f"Missing folder for staging: '{dir_staging}'"  # noqa: S101
+
+        import shutil
+
+        for pkg in os.listdir(dir_local):
+            if not os.path.isfile(pkg):
+                continue
+            if pkg in os.listdir(dir_cache):
+                continue
+            logging.info(f"Moving '{pkg}' to staging...")
+            shutil.move(os.path.join(dir_cache, pkg), os.path.join(dir_staging, pkg))
 
 
 if __name__ == "__main__":

.github/workflows/_focus-diff.yml

@@ -27,7 +27,6 @@ jobs:
           set -e
           echo $PR_NUMBER
           pip install -q -U packaging fire pyGithub pyopenssl
-          # python .github/assistant.py changed-domains $PR_NUMBER
           echo "focus=$(python .github/assistant.py changed-domains $PR_NUMBER)" >> $GITHUB_OUTPUT
 
       - run: echo "${{ steps.diff-domains.outputs.focus }}"