Test against stable pytorch instead of nightly

.github/workflows/build-cuda.yml

@@ -17,7 +17,7 @@ jobs:
         include:
           - name: 4xlargegpu
             runs-on: linux.g5.4xlarge.nvidia.gpu
-            torch-spec: '--pre torch --index-url https://download.pytorch.org/whl/nightly/cu126'
+            torch-spec: 'torch'
             gpu-arch-type: "cuda"
             gpu-arch-version: "12.6"
     with:

.github/workflows/publish_release.yml

@@ -21,7 +21,7 @@ jobs:
         include:
           - name: 4xlarge
             runs-on: linux.g5.4xlarge.nvidia.gpu
-            torch-spec: '--pre torch --index-url https://download.pytorch.org/whl/nightly/cu128'
+            torch-spec: 'torch'
             gpu-arch-type: "cuda"
             gpu-arch-version: "12.8"
     uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main

.github/workflows/test-gpu-python.yml

@@ -22,7 +22,7 @@ jobs:
         include:
           - name: 4xlargegpu
             runs-on: linux.g5.4xlarge.nvidia.gpu
-            torch-spec: '--pre torch --index-url https://download.pytorch.org/whl/nightly/cu126'
+            torch-spec: 'torch'
             gpu-arch-type: "cuda"
             gpu-arch-version: "12.6"
     with:

.github/workflows/test-gpu-rust.yml

@@ -22,7 +22,7 @@ jobs:
         include:
           - name: 4xlargegpu
             runs-on: linux.g5.4xlarge.nvidia.gpu
-            torch-spec: '--pre torch --index-url https://download.pytorch.org/whl/nightly/cu126'
+            torch-spec: 'torch'
             gpu-arch-type: "cuda"
             gpu-arch-version: "12.6"
     with:

.github/workflows/wheels.yml

@@ -21,7 +21,7 @@ jobs:
         include:
           - name: 4xlarge
             runs-on: linux.g5.4xlarge.nvidia.gpu
-            torch-spec: '--pre torch --index-url https://download.pytorch.org/whl/nightly/cu126'
+            torch-spec: 'torch'
             gpu-arch-type: "cuda"
             gpu-arch-version: "12.6"
     uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main

README.md

@@ -102,7 +102,6 @@ conda update -n monarchenv --all -c conda-forge -y
 sudo dnf install -y libibverbs rdma-core libmlx5 libibverbs-devel rdma-core-devel
 
 # Install build dependencies
-pip install --pre torch --index-url https://download.pytorch.org/whl/nightly/cu128
 pip install -r build-requirements.txt
 # Install test dependencies
 pip install -r python/tests/requirements.txt
@@ -142,7 +141,6 @@ export CXX=clang++
 sudo apt install -y cuda-toolkit-12-8 cuda-12-8
 
 # Install build dependencies
-pip install --pre torch --index-url https://download.pytorch.org/whl/nightly/cu128
 pip install -r build-requirements.txt
 # Install test dependencies
 pip install -r python/tests/requirements.txt
@@ -180,7 +178,6 @@ rustup toolchain install nightly
 rustup default nightly
 
 # Install build dependencies
-pip install --pre torch --index-url https://download.pytorch.org/whl/nightly/cpu
 pip install -r build-requirements.txt
 # Install test dependencies
 pip install -r python/tests/requirements.txt

python/monarch/gradient/_gradient_generator.cpp

@@ -426,7 +426,7 @@ struct GradientGenerator {
     DEBUG_PRINT(
         "// add: " << node->node->name()
                    << ", input_nr=" << static_cast<int>(input_nr) << "\n");
-#if TORCH_VERSION_NEWER_THAN(2, 8, 0)
+#if TORCH_VERSION_NEWER_THAN(2, 10, 0)
     realInputBuffer(node).add(
         input_nr,
         check_and_reduce(node->node, input_nr, std::move(t)),

scripts/common-setup.sh

@@ -70,7 +70,7 @@ setup_tensor_engine() {
 # Install PyTorch with C++ development headers (libtorch) for Rust compilation
 setup_pytorch_with_headers() {
     local gpu_arch_version=${1:-"12.6"}
-    local torch_spec=${2:-"--pre torch --index-url https://download.pytorch.org/whl/nightly/cu126"}
+    local torch_spec=${2:-"torch"}
 
     echo "Setting up PyTorch with C++ headers (GPU arch: ${gpu_arch_version})..."