handle older PyTorch versions

.github/workflows/main.yml

@@ -9,54 +9,82 @@ on:
 
 jobs:
 
-  setup-dev-env:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
-      - uses: ./.github/actions/setup-env
-        with:
-          environment: dev
-
   check:
     runs-on: ubuntu-latest
-    needs: setup-dev-env
     steps:
       - uses: actions/checkout@v4
-      - uses: ./.github/actions/setup-env
+      - uses: prefix-dev/setup-[email protected]
         with:
-          environment: dev
+          pixi-version: v0.27.1
+          frozen: true
+          cache: true
+          environments: dev
+          activate-environment: dev
       - run: ruff check
         if: success() || failure()
       - run: ruff format --check
         if: success() || failure()
 
   ##
 
-  setup-default-env:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
-      - uses: ./.github/actions/setup-env
-        with:
-          environment: default
-
   typecheck:
     runs-on: ubuntu-latest
-    needs: setup-default-env
     steps:
       - uses: actions/checkout@v4
-      - uses: ./.github/actions/setup-env
+      - uses: prefix-dev/setup-[email protected]
         with:
-          environment: default
+          pixi-version: v0.27.1
+          frozen: true
+          cache: true
+          environments: default
+          activate-environment: default
       - run: pyright
         if: success() || failure()
 
+  ##
+
+  get-pytorch-versions:
+    runs-on: ubuntu-latest
+    outputs:
+      versions: ${{ steps.get-pytorch-versions.outputs.versions }}
+    steps:
+      - name: Get PyTorch versions
+        id: get-pytorch-versions
+        run: |
+          VERSIONS=$(
+            curl -s https://pypi.org/pypi/torch/json | jq -r '.releases | keys[]' |
+            # remove versions <2.0; strip "patch" from versions
+            grep -v '^1\.' | grep -E '\.[0]+$' | sort -V | sed 's/\.0$//' |
+            # to JSON array
+            jq -R . | jq -sc .
+          )
+          echo "versions=$VERSIONS" >> $GITHUB_OUTPUT
+          # e.g. ["2.0","2.1","2.2","2.3","2.4"]
+
   test:
     runs-on: ubuntu-latest
-    needs: setup-default-env
+    needs: get-pytorch-versions
+    strategy: 
+      fail-fast: false
+      matrix:
+        python: ["3.8", "3.9", "3.10", "3.11", "3.12"] 
+        pytorch: ${{fromJson(needs.get-pytorch-versions.outputs.versions)}}
+
     steps:
       - uses: actions/checkout@v4
-      - uses: ./.github/actions/setup-env
+
+      - name: Inject (python / pytorch) test deps into pixi.toml / pyproject.toml
+        run: |
+          sed -i 's/python = "3.8.1"/python = "${{ matrix.python }}.*"/' pixi.toml
+          sed -i 's/torch>=2.0.0/torch~=${{ matrix.pytorch }}/' pyproject.toml
+
+      - uses: prefix-dev/[email protected]
         with:
-          environment: default
+          pixi-version: v0.27.1
+          locked: false
+          frozen: false
+          cache: false
+          environments: default
+          activate-environment: default
+
       - run: pytest tests

docs/source/advanced.rst

@@ -35,4 +35,8 @@ The :mod:`torchrunx.launch` ``env_vars`` argument allows the user to specify whi
 :mod:`torchrunx.launch` also accepts the ``env_file`` argument, which is designed to expose more advanced environmental configuration to the user. When a file is provided as this argument, the launcher will source the file on each node before executing the agent. This allows for custom bash scripts to be provided in the environmental variables, and allows for node-specific environmental variables to be set.
 
 ..
-    TODO: example env_file
+    TODO: example env_file
+
+Numpy >= 2.0
+------------
+only supported if `torch>=2.3`

examples/slurm_poc.py

@@ -21,7 +21,7 @@ def test_launch():
     print("PASS")
 
 
-def simple_matmul(test):
+def simple_matmul():
     rank = int(os.environ["RANK"])
     local_rank = int(os.environ["LOCAL_RANK"])
     device = torch.device(local_rank) if torch.cuda.is_available() else torch.device("cpu")
@@ -35,7 +35,6 @@ def simple_matmul(test):
 
     i = torch.rand((500, 100), device=device)  # batch, dim
     o = torch.matmul(i, w)
-    print(test)
     dist.all_reduce(o, op=dist.ReduceOp.SUM)
     print(i)
     return o.detach().cpu()