Merge remote-tracking branch 'origin/main' into tom/drop-old-style-di…

…stributed
Lightning-AI · Jan 13, 2025 · a477a7c · a477a7c
2 parents 0deac3c + f56f71a
commit a477a7c
Show file tree

Hide file tree

Showing 2 changed files with 68 additions and 6 deletions.
diff --git a/.azure/gpu-tests.yml b/.azure/gpu-tests.yml
@@ -16,16 +16,34 @@ jobs:
     strategy:
       matrix:
         # CUDA 12.1
-        "ubuntu22.04 | cuda 12.1 | python 3.10 | torch 2.5.1 | regular":
+        "ubuntu22.04 | cuda 12.1 | python 3.10 | torch 2.5.1 | main":
           docker-image: "ubuntu22.04-cuda12.1.1-cudnn-fe1.5.2-py3.10-pt_2.5.1-dev"
           CUDA_VERSION_MM: "121"
+          testing: "main"
+        "ubuntu22.04 | cuda 12.1 | python 3.10 | torch 2.5.1 | ops":
+          docker-image: "ubuntu22.04-cuda12.1.1-cudnn-fe1.5.2-py3.10-pt_2.5.1-dev"
+          CUDA_VERSION_MM: "121"
+          testing: "ops"
+        "ubuntu22.04 | cuda 12.1 | python 3.10 | torch 2.5.1 | grads":
+          docker-image: "ubuntu22.04-cuda12.1.1-cudnn-fe1.5.2-py3.10-pt_2.5.1-dev"
+          CUDA_VERSION_MM: "121"
+          testing: "grads"
         "ubuntu22.04 | cuda 12.1 | python 3.10 | torch 2.5.1 | distributed":
           docker-image: "ubuntu22.04-cuda12.1.1-cudnn-fe1.5.2-py3.10-pt_2.5.1-dev"
           CUDA_VERSION_MM: "121"
           testing: "distributed"
-        "ubuntu22.04 | cuda 12.1 | python 3.10 | torch-nightly | regular":
+        "ubuntu22.04 | cuda 12.1 | python 3.10 | torch-nightly | main":
+          docker-image: "ubuntu22.04-cuda12.1.1-cudnn-fe1.5.2-py3.10-pt_main-dev"
+          CUDA_VERSION_MM: "121"
+          testing: "main"
+        "ubuntu22.04 | cuda 12.1 | python 3.10 | torch-nightly | ops":
           docker-image: "ubuntu22.04-cuda12.1.1-cudnn-fe1.5.2-py3.10-pt_main-dev"
           CUDA_VERSION_MM: "121"
+          testing: "ops"
+        "ubuntu22.04 | cuda 12.1 | python 3.10 | torch-nightly | grads":
+          docker-image: "ubuntu22.04-cuda12.1.1-cudnn-fe1.5.2-py3.10-pt_main-dev"
+          CUDA_VERSION_MM: "121"
+          testing: "grads"
         "ubuntu22.04 | cuda 12.1 | python 3.10 | torch-nightly | distributed":
           docker-image: "ubuntu22.04-cuda12.1.1-cudnn-fe1.5.2-py3.10-pt_main-dev"
           CUDA_VERSION_MM: "121"
@@ -88,16 +106,17 @@ jobs:
               --durations=250 \
               --timeout=240 \
               --numprocesses=9 \
-              --ignore=thunder/tests/distributed --ignore=thunder/tests/test_networks.py
+              --ignore=thunder/tests/distributed --ignore=thunder/tests/test_networks.py \
+              --ignore=thunder/tests/test_ops.py --ignore=thunder/tests/test_grad.py
           # compile coverage results
           python -m coverage report
           python -m coverage xml
           # upload to codecov
           ./codecov --token=$(CODECOV_TOKEN) --commit=$(Build.SourceVersion) \
             --flags=gpu,pytest,regular --name="GPU-coverage" --env=linux,azure
-        condition: ne(variables['testing'], 'distributed')
+        condition: eq(variables['testing'], 'main')
         timeoutInMinutes: "40"
-        displayName: "Testing: regular"
+        displayName: "Testing: main"
 
       - bash: |
           set -ex
@@ -115,10 +134,52 @@ jobs:
           # upload to codecov
           ./codecov --token=$(CODECOV_TOKEN) --commit=$(Build.SourceVersion) \
             --flags=gpu,pytest,networks --name="GPU-coverage" --env=linux,azure
-        condition: ne(variables['testing'], 'distributed')
+        condition: eq(variables['testing'], 'main')
         timeoutInMinutes: "15"
         displayName: "Testing: networks"
 
+      - bash: |
+          set -ex
+          export CUDA_LAUNCH_BLOCKING=1
+          coverage run --source thunder -m \
+            pytest thunder/tests/test_ops.py \
+              -m "not standalone" \
+              -v --datefmt="%Y%m%d-%H:%M:%S.%f" \
+              --random-order-seed=42 \
+              --durations=250 \
+              --timeout=240 \
+              --numprocesses=9
+          # compile coverage results
+          python -m coverage report
+          python -m coverage xml
+          # upload to codecov
+          ./codecov --token=$(CODECOV_TOKEN) --commit=$(Build.SourceVersion) \
+            --flags=gpu,pytest,regular --name="GPU-coverage" --env=linux,azure
+        condition: eq(variables['testing'], 'ops')
+        timeoutInMinutes: "40"
+        displayName: "Testing: ops"
+
+      - bash: |
+          set -ex
+          export CUDA_LAUNCH_BLOCKING=1
+          coverage run --source thunder -m \
+            pytest thunder/tests/test_grad.py \
+              -m "not standalone" \
+              -v --datefmt="%Y%m%d-%H:%M:%S.%f" \
+              --random-order-seed=42 \
+              --durations=250 \
+              --timeout=240 \
+              --numprocesses=9
+          # compile coverage results
+          python -m coverage report
+          python -m coverage xml
+          # upload to codecov
+          ./codecov --token=$(CODECOV_TOKEN) --commit=$(Build.SourceVersion) \
+            --flags=gpu,pytest,regular --name="GPU-coverage" --env=linux,azure
+        condition: eq(variables['testing'], 'grads')
+        timeoutInMinutes: "40"
+        displayName: "Testing: grads"
+
       - bash: |
           set -ex
           # run all found tests in given past as standalone

diff --git a/docs/.readthedocs.yaml b/docs/.readthedocs.yaml
@@ -9,6 +9,7 @@ version: 2
 # reference: https://docs.readthedocs.io/en/stable/config-file/v2.html#sphinx
 sphinx:
   fail_on_warning: true
+  configuration: docs/conf.py
 
 build:
   os: "ubuntu-22.04"