Skip to content
Closed
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 3 additions & 1 deletion .github/container/Dockerfile.jax
Original file line number Diff line number Diff line change
Expand Up @@ -39,6 +39,7 @@ ARG EXTRA_BAZEL_TARGETS
ARG EXTRA_BUILD_JAX_ARGS
ARG GIT_USER_NAME
ARG GIT_USER_EMAIL
ARG NVTE_SKIP_SUBMODULE_CHECKS_DURING_BUILD=1

RUN --mount=type=ssh \
--mount=type=secret,id=SSH_KNOWN_HOSTS,target=/root/.ssh/known_hosts \
Expand All @@ -47,7 +48,7 @@ RUN --mount=type=ssh \
git-clone.sh ${URLREF_XLA} ${SRC_PATH_XLA}
EOF

ADD build-jax.sh build-te.sh local_cuda_arch /usr/local/bin/
ADD build-jax.sh build-te.sh local_cuda_arch /usr/local/bin/
# Install bazelisk
RUN ARCH="$(dpkg --print-architecture)" && \
wget -O /usr/local/bin/bazel https://github.com/bazelbuild/bazelisk/releases/latest/download/bazelisk-linux-${ARCH} && \
Expand All @@ -70,6 +71,7 @@ RUN mkdir -p /builder/extra-targets/{bin,python} && \
RUN --mount=type=ssh \
--mount=type=secret,id=SSH_KNOWN_HOSTS,target=/root/.ssh/known_hosts \
git-clone.sh ${URLREF_TRANSFORMER_ENGINE} ${SRC_PATH_TRANSFORMER_ENGINE}
RUN pushd ${SRC_PATH_TRANSFORMER_ENGINE}/3rdparty/cudnn-frontend && git fetch && git checkout f06a0766b664b73860c48f7aea2b6b09499cea8d && popd
# Populate ${SRC_PATH_TRANSFORMER_ENGINE}/dist with [a] .whl file(s); --no-install
# because (a) this is the builder stage, and (b) pip-finalize.sh does the install
RUN build-te.sh \
Expand Down
23 changes: 23 additions & 0 deletions .github/container/cudnn_frontend.patch
Original file line number Diff line number Diff line change
@@ -0,0 +1,23 @@
diff --git a/include/cudnn_frontend_shim.h b/include/cudnn_frontend_shim.h
index ea838d3..c8f9de7 100644
--- a/include/cudnn_frontend_shim.h
+++ b/include/cudnn_frontend_shim.h
@@ -96,12 +96,12 @@ load_cudart_so() {
const char *error = reinterpret_cast<const char *>(dlerror());

if (handle && !error) {
- if (lib_handle) {
- // Already loaded one -> multiple found
- dlclose(handle);
- throw std::runtime_error("Multiple libcudart libraries found: " + std::string(libs[loaded_index]) +
- " and " + std::string(libs[i]));
- }
+ // if (lib_handle) {
+ // // Already loaded one -> multiple found
+ // dlclose(handle);
+ // throw std::runtime_error("Multiple libcudart libraries found: " + std::string(libs[loaded_index]) +
+ // " and " + std::string(libs[i]));
+ // }
lib_handle = handle;
loaded_index = static_cast<int>(i);
}
4 changes: 3 additions & 1 deletion .github/workflows/_test_maxtext_gke_xpk.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -45,8 +45,10 @@ jobs:
with:
IMAGE: ${{ inputs.MAXTEXT_IMAGE }}
NAME: ${{ matrix.test }}
ENVS: |
CUDNN_FRONTEND_CUDART_LIB_NAME=libcudart.so.13;
COMMAND: |
source \${NCCL_LIB_DIR}/nccl-env-profile.sh;
NCCL_LIB_DIR=\${NCCL_LIB_DIR} . \${NCCL_LIB_DIR}/nccl-env-profile.sh;
nsys-jax --capture-range=cudaProfilerApi \
--capture-range-end=stop \
-o /opt/output/profile.zip \
Expand Down
Loading