25.10 NGC release testing

.github/actions/gke-xpk/action.yml

@@ -247,6 +247,7 @@ runs:
 
         if [ $? -ne 0 ]; then
           echo "The JobSet ${WORKLOAD_NAME} on ${{ inputs.GKE_CLUSTER }} did not complete as expected "
+          echo "XPK_EXIT_CODE=1" >> ${GITHUB_ENV}
           exit 1
         fi
 
@@ -262,11 +263,12 @@ runs:
         ALL_EXIT_CODES=$(( ALL_EXIT_CODES + POD_EXIT_CODE ))
       done
 
+      echo "XPK_EXIT_CODE=${ALL_EXIT_CODES}" >> ${GITHUB_ENV}
       if [ ${ALL_EXIT_CODES} -gt 0 ]; then
         exit 1
       fi
       exit 0
-  
+
   - name: Clean up JobSet from cluster
     shell: bash -x -u {0}
     if: ${{ always() }}
@@ -291,3 +293,38 @@ runs:
     if: ${{ always() }}
     run: |
       sudo rm -rf ${WORKLOAD_NAME}
+
+  - name: Generate sitrep
+    id: sitrep
+    shell: bash -x -e {0}
+    if: ${{ always() }}
+    run: |
+      source .github/workflows/scripts/to_json.sh
+      badge_label="${{ matrix.test }}"
+
+      summary="${{ inputs.WORKLOAD_NAME_PREFIX }}"
+      outcome=success
+      badge_label="${{ inputs.WORKLOAD_NAME_PREFIX }}"
+      badge_color=brightgreen
+
+      if [ "${XPK_EXIT_CODE}" -gt 0 ]; then
+        badge_color=red
+        outcome=failed
+        summary+=": fail"
+      else
+        summary+=": pass"
+      fi
+
+      to_json summary \
+              badge_label \
+              badge_color \
+              outcome | \
+      tee sitrep.json
+
+  - name: Upload sitrep to GitHub Actions from runner
+    if: ${{ always() }}
+    uses: actions/upload-artifact@v4
+    with:
+      name: ${{ inputs.WORKLOAD_NAME_PREFIX }}-sitrep
+      path: |
+        sitrep.json

.github/container/Dockerfile.jax

@@ -64,6 +64,7 @@ RUN mkdir -p /builder/extra-targets && \
     --src-path-xla ${SRC_PATH_XLA} \
     --sm all \
     --clean \
+    --release \
     ${EXTRA_BUILD_JAX_ARGS}
 
 ## Transformer engine: check out source and build wheel

.github/container/build-jax.sh

@@ -83,7 +83,7 @@ INSTALL=1
 SRC_PATH_JAX="/opt/jax"
 SRC_PATH_XLA="/opt/xla"
 
-args=$(getopt -o h --long bazel-cache:,bazel-cache-namespace:,build-param:,build-path-jaxlib:,clean,cpu-arch:,debug,extra-targets:,extra-target-dest:,no-clean,clean-only,help,install,no-install,src-path-jax:,src-path-xla:,sm: -- "$@")
+args=$(getopt -o h,r --long bazel-cache:,bazel-cache-namespace:,build-param:,build-path-jaxlib:,clean,release,cpu-arch:,debug,extra-targets:,extra-target-dest:,no-clean,clean-only,help,install,no-install,src-path-jax:,src-path-xla:,sm: -- "$@")
 if [[ $? -ne 0 ]]; then
     exit 1
 fi
@@ -135,6 +135,10 @@ while [ : ]; do
             EXTRA_TARGET_DEST="$2"
             shift 2
             ;;
+        -r | --release)
+            IS_RELEASE=1
+            shift 1
+            ;;
         -h | --help)
             usage 1
             ;;
@@ -225,6 +229,7 @@ print_var INSTALL
 print_var PYTHON_VERSION
 print_var SRC_PATH_JAX
 print_var SRC_PATH_XLA
+print_var IS_RELEASE
 
 echo "=================================================="
 
@@ -268,6 +273,12 @@ for component in jaxlib "jax-cuda${CUDA_MAJOR_VERSION}-pjrt" "jax-cuda${CUDA_MAJ
     # version, so nvidia-*-cu12 wheels disappear from the lock file
     sed -i "s|^${component}.*$|${component} @ file://${BUILD_PATH_JAXLIB}/${component//-/_}|" build/requirements.in
 done
+
+if [[ "${IS_RELEASE}" == "1" ]]; then
+    jaxlib_version=$(pip show jaxlib | grep Version | tr ':' '\n' | tail -1)
+    sed -i "s|      f'jaxlib >={_minimum_jaxlib_version}, <={_jax_version}',|      f'jaxlib>=0.5.0',|" /opt/jax/setup.py
+fi
+
 # Bazel args to avoid cache invalidation
 BAZEL_ARGS=(
     --config=cuda_libraries_from_stubs

.github/container/build-te.sh

@@ -103,6 +103,12 @@ if [[ "$SM" == "all" ]]; then
     SM_LIST=$(default_compute_capabilities)
 elif [[ "$SM" == "local" ]]; then
     SM_LIST=$("${SCRIPT_DIR}/local_cuda_arch")
+    if [[ -z "${SM_LIST}" ]]; then
+        echo "Could not determine the local GPU architecture."
+        echo "You should pass --sm when compiling on a machine without GPUs."
+        nvidia-smi || true
+        exit 1
+    fi
 else
     SM_LIST=${SM}
 fi
@@ -131,8 +137,19 @@ export NVTE_FRAMEWORK=jax
 export XLA_HOME=${SRC_PATH_XLA}
 
 pushd ${SRC_PATH_TE}
-# Install required packages that were removed in https://github.com/NVIDIA/TransformerEngine/pull/1852
-pip install "pybind11[global]"
+# Install some build dependencies, but avoid installing everything
+# (jax, torch, ...) because we do not want to pull in a released version of
+# JAX, or the wheel-based installation of CUDA. Note that when we build TE as
+# part of building the JAX containers, JAX and XLA are not yet installed.
+python - << EOF
+import subprocess, sys, tomllib
+with open("pyproject.toml", "rb") as ifile:
+    data = tomllib.load(ifile)
+subprocess.run(
+    [sys.executable, "-m", "pip", "install"]
+    + [r for r in data["build-system"]["requires"]
+       if r.startswith("nvidia-mathdx") or r.startswith("pybind11")])
+EOF
 
 # The wheel filename includes the TE commit; if this has changed since the last
 # incremental build then we would end up with multiple wheels.

.github/container/git-clone.sh

@@ -77,6 +77,13 @@ pushd ${DESTINATION}
 git checkout ${GIT_REF}
 COMMIT_SHA=$(git rev-parse HEAD)
 git submodule update --init --recursive
+if [[ "${GIT_REPO}" == *"gitlab"* ]]; then
+  git remote remove origin
+  if grep -q -r gitlab-ci-token .git; then
+    grep -r gitlab-ci-token .git | awk -F: '{print $1}' | xargs rm -f
+  fi
+  git branch -D main
+fi
 popd
 
 ## update the manifest file

.github/container/test-jax.sh

@@ -119,7 +119,15 @@ fi
 
 readarray -t GPU_MEMORIES < <(nvidia-smi --query-gpu=memory.total --format=csv,noheader)
 NGPUS="${#GPU_MEMORIES[@]}"
-GPU_MEMORIES_MIB=("${GPU_MEMORIES[@]/ MiB/}")
+if [[ " ${GPU_MEMORIES[*]} " =~ [[:space:]]\[N/A\][[:space:]] ]]; then
+    # On iGPU devices, nvidia-smi reports [N/A] GPU memory; use the system
+    # memory size instead to estimate what each GPU can use
+    SYSTEM_MEMORY_MIB=$(grep MemTotal /proc/meminfo | awk '{print $2 / 1024}')
+    declare -a GPU_MEMORIES_MIB
+    for (( i = 0; i < NGPUS; i++ )); do GPU_MEMORIES_MIB+=($(( SYSTEM_MEMORY_MIB / NGPUS ))); done
+else
+    GPU_MEMORIES_MIB=("${GPU_MEMORIES[@]/ MiB/}")
+fi
 
 FLAGS=()
 

.github/eks-workflow-files/maxtext-job.yaml

@@ -0,0 +1,120 @@
+apiVersion: v1
+kind: Service
+metadata:
+  name: PLACEHOLDER
+spec:
+  clusterIP: None # clusterIP must be None to create a headless service
+  selector:
+    job-name: PLACEHOLDER # must match Job name
+---
+apiVersion: batch/v1
+kind: Job
+metadata:
+  name: PLACEHOLDER
+  labels:
+    kueue.x-k8s.io/queue-name: p5-queue
+spec:
+  completions: 2 # number of nodes
+  parallelism: 2 # number of nodes
+  completionMode: Indexed
+  backoffLimitPerIndex: 0 # max failures per index
+  maxFailedIndexes:     0 # all indices must succeed
+  template:
+    spec:
+      subdomain: PLACEHOLDER # has to match Service name
+      restartPolicy: Never
+      imagePullSecrets:
+        - name: PLACEHOLDER
+      containers:
+        - name: maxtext
+          image: PLACEHOLDER
+          ports:
+            - containerPort: 3389
+          command:
+            - bash
+            - -c
+            # The logging logic: stream stdout/stderr from the 0th process inside this pod,
+            # record all of the processes' stdout/stderr + the INFO-level NCCL logs to file
+            - |
+              export SERVICE_NAME=$0
+              export JOB_NAME=$1
+              cat >each-process.sh <<'EOL'
+              export JAX_COORDINATOR_IP=${JOB_NAME}-0.${SERVICE_NAME}
+              export JAX_COORDINATOR_PORT=3389
+              export NNODES=16 # actually #processes == #GPUs
+              export NODE_RANK=$((JOB_COMPLETION_INDEX*8 + LOCAL_RANK))
+              export JAX_LOCAL_DEVICE_IDS=$LOCAL_RANK
+              export NCCL_DEBUG=INFO
+              export NCCL_DEBUG_FILE=/opt/output/nccl.$NODE_RANK.log
+              [[ $LOCAL_RANK == 0 ]] && console="/dev/stdout" || console="/dev/null"
+              nsys-jax \
+                --capture-range=cudaProfilerApi \
+                --capture-range-end=stop \
+                -o /opt/output/profile.$NODE_RANK.zip \
+                -- \
+                test-maxtext.sh \
+                -n 2 \
+                -b 2 \
+                --model-name=llama2-7b \
+                --attn-type=cudnn_flash_te \
+                --remat-policy=minimal_flash \
+                --steps=20 \
+                --fsdp=16 \
+                -a "scan_layers=false \
+                    max_target_length=4096 \
+                    use_iota_embed=true \
+                    logits_dot_in_fp32=false \
+                    profiler=nsys \
+                    skip_first_n_steps_for_profiler=3 \
+                    profiler_steps=8" \
+                |& tee /opt/output/output.$NODE_RANK.log >"${console}"
+              code=$?
+              # Should run even on failure
+              cat /opt/output/nccl.$NODE_RANK.log >"${console}"
+              exit $code
+              EOL
+              # TODO: upgrade parallel-launch to return a failure code as soon as any
+              #       of its children do (it already does this eventually, but it could
+              #       be slow)
+              parallel-launch LOCAL_RANK 8 bash each-process.sh
+              code=$?
+              # Should run even on failure
+              touch /opt/output/.done
+              exit $code
+            - PLACEHOLDER
+            - PLACEHOLDER
+          resources:
+            limits:
+              nvidia.com/gpu: 8
+              vpc.amazonaws.com/efa: 32
+          volumeMounts:
+            - mountPath: /dev/shm
+              name: shmem
+            - mountPath: /opt/output
+              name: output
+        - name: upload
+          image: amazon/aws-cli
+          command:
+            - bash
+            - -c
+            - |
+              JOB_NAME="$0"
+              while [[ ! -f /opt/output/.done ]]; do
+                sleep 1
+              done
+              rm /opt/output/.done
+              aws s3 cp \
+                --recursive \
+                /opt/output \
+                "s3://jax-toolbox-eks-output/${JOB_NAME}/"
+            - PLACEHOLDER
+          volumeMounts:
+            - mountPath: /opt/output
+              name: output
+      volumes:
+        - name: output
+          emptyDir: {}
+        - name: shmem
+          emptyDir:
+            medium: Memory
+            sizeLimit: 16Gi

.github/workflows/_test_maxtext_gke_xpk.yaml

@@ -26,14 +26,14 @@ jobs:
     steps:
     - uses: actions/checkout@v4
 
-    - name: Login to GitHub Container Registry
+    - name: Login to nvcr.io Container Registry
       uses: docker/login-action@v3
       with:
-        registry: ghcr.io
-        username: ${{ github.repository_owner }}
-        password: ${{ secrets.GITHUB_TOKEN }}
+        registry: nvcr.io
+        username: $oauthtoken
+        password: ${{ secrets.NVCR_TOKEN }}
 
-    - name: K8s GHCR store and delete token
+    - name: K8s store and delete token
       id: store-token
       uses: ./.github/actions/store-delete-k8s-ghcr