aws · sirutBuasai · Jul 14, 2026 · Jul 14, 2026 · Jul 14, 2026 · Jul 14, 2026
diff --git a/.github/config/image/lambda/base-preview.yml b/.github/config/image/lambda/base-preview.yml
@@ -1,6 +1,6 @@
 image:
   name: "lambda-base-preview"
-  description: "Lambda base GPU image with thread-mode RIC preview"
+  description: "Lambda base GPU image with multi-mode concurrency RIC"
 
 metadata:
   framework: "lambda"
@@ -19,6 +19,8 @@ build:
   cuda_version: "12.9.1"
   dlc_major_version: "1"
   dlc_minor_version: "0"
+  awslambdaric_version: "3.1.1"
+  awslambdaric_release: "2.0.0.0"
 
 release:
   release: true

diff --git a/.github/config/image/lambda/cupy-preview.yml b/.github/config/image/lambda/cupy-preview.yml
@@ -1,6 +1,6 @@
 image:
   name: "lambda-cupy-preview"
-  description: "Lambda CuPy GPU image with thread-mode RIC preview"
+  description: "Lambda CuPy GPU image with multi-mode concurrency RIC"
 
 metadata:
   framework: "lambda"
@@ -19,6 +19,8 @@ build:
   cuda_version: "12.9.1"
   dlc_major_version: "1"
   dlc_minor_version: "0"
+  awslambdaric_version: "3.1.1"
+  awslambdaric_release: "2.0.0.0"
 
 release:
   release: true

diff --git a/.github/config/image/lambda/pytorch-preview.yml b/.github/config/image/lambda/pytorch-preview.yml
@@ -1,6 +1,6 @@
 image:
   name: "lambda-pytorch-preview"
-  description: "Lambda PyTorch GPU image with thread-mode RIC preview"
+  description: "Lambda PyTorch GPU image with multi-mode concurrency RIC"
 
 metadata:
   framework: "lambda"
@@ -19,6 +19,8 @@ build:
   cuda_version: "12.9.1"
   dlc_major_version: "1"
   dlc_minor_version: "0"
+  awslambdaric_version: "3.1.1"
+  awslambdaric_release: "2.0.0.0"
 
 release:
   release: true

diff --git a/.github/workflows/lambda.pipeline.yml b/.github/workflows/lambda.pipeline.yml
@@ -158,6 +158,12 @@ jobs:
               TESTS="$COMMON /test/unit/test_imports_base.py" ;;
           esac
           echo "tests=${TESTS}" >> $GITHUB_OUTPUT
+          # Multi-mode RIC targets carry a runtime concurrency-mode knob.
+          if [[ "$TARGET" == *preview* ]]; then
+            echo "multimode=true" >> $GITHUB_OUTPUT
+          else
+            echo "multimode=false" >> $GITHUB_OUTPUT
+          fi
 
       - name: Run GPU validation tests
         if: steps.check.outputs.exists == 'true'
@@ -167,9 +173,19 @@ jobs:
             ${{ steps.image.outputs.image-uri }} \
             -m pytest ${{ steps.config.outputs.tests }} -v
 
-      - name: Skip notice
-        if: steps.check.outputs.exists != 'true'
-        run: echo "Image not found in ECR — skipping tests (first release scenario)"
+      - name: Verify runtime concurrency-mode override
+        if: steps.check.outputs.exists == 'true' && steps.config.outputs.multimode == 'true'
+        run: |
+          for mode in process thread hybrid; do
+            echo "::group::override AWS_LAMBDA_CONCURRENCY_MODE=${mode}"
+            docker run --rm --gpus all --entrypoint python \
+              -e AWS_LAMBDA_CONCURRENCY_MODE=${mode} \
+              -e EXPECTED_CONCURRENCY_MODE=${mode} \
+              -v $(pwd)/test/lambda:/test \
+              ${{ steps.image.outputs.image-uri }} \
+              -m pytest /test/unit/test_lambda_runtime.py::test_ric_resolves_container_env -v
+            echo "::endgroup::"
+          done
 
   release-gate:
     if: >-

diff --git a/docker/lambda/Dockerfile b/docker/lambda/Dockerfile
@@ -144,21 +144,21 @@ RUN --mount=type=cache,target=/root/.cache/uv \
 # ============================================================
 FROM nvidia/cuda:${CUDA_VERSION}-runtime-amzn2023 as oss-base-py3
 COPY --from=builder-base-py3 /var/lang /var/lang
-COPY ./scripts/common/setup_oss_compliance.sh /tmp/setup_oss_compliance.sh
+COPY ./scripts/docker/common/setup_oss_compliance.sh /tmp/setup_oss_compliance.sh
 RUN PATH="/var/lang/bin:$PATH" bash /tmp/setup_oss_compliance.sh python3 \
   && touch /root/THIRD_PARTY_SOURCE_CODE_URLS
 
 FROM nvidia/cuda:${CUDA_VERSION}-runtime-amzn2023 as oss-cupy-py3
 COPY --from=builder-cupy-py3 /var/lang /var/lang
-COPY ./scripts/common/setup_oss_compliance.sh /tmp/setup_oss_compliance.sh
+COPY ./scripts/docker/common/setup_oss_compliance.sh /tmp/setup_oss_compliance.sh
 RUN PATH="/var/lang/bin:$PATH" bash /tmp/setup_oss_compliance.sh python3 \
   && touch /root/THIRD_PARTY_SOURCE_CODE_URLS
 
 FROM nvidia/cuda:${CUDA_VERSION}-runtime-amzn2023 as oss-pytorch-py3
 ARG FFMPEG_VERSION
 COPY --from=builder-pytorch-py3 /var/lang /var/lang
 COPY --from=lambda-python /lib64/libssl.so.3 /lib64/libcrypto.so.3 /lib64/
-COPY ./scripts/common/setup_oss_compliance.sh /tmp/setup_oss_compliance.sh
+COPY ./scripts/docker/common/setup_oss_compliance.sh /tmp/setup_oss_compliance.sh
 RUN PATH="/var/lang/bin:$PATH" bash /tmp/setup_oss_compliance.sh python3 \
   && printf '\n** FFmpeg; version n%s -- https://github.com/FFmpeg/FFmpeg/archive/refs/tags/n%s.tar.gz\n' "${FFMPEG_VERSION}" "${FFMPEG_VERSION}" >> /root/BUILD_FROM_SOURCE_PACKAGES_LICENCES \
   && printf '\nFFmpeg n%s https://github.com/FFmpeg/FFmpeg/archive/refs/tags/n%s.tar.gz\n' "${FFMPEG_VERSION}" "${FFMPEG_VERSION}" >> /root/THIRD_PARTY_SOURCE_CODE_URLS \
@@ -192,8 +192,8 @@ ENV PATH="/var/lang/bin:$PATH" \
   LAMBDA_RUNTIME_DIR=/var/runtime \
   LANG=en_US.UTF-8 \
   TZ=:/etc/localtime
-COPY ./scripts/telemetry/deep_learning_container.py /usr/local/bin/deep_learning_container.py
-COPY ./scripts/telemetry/bash_telemetry.sh.template /tmp/bash_telemetry.sh.template
+COPY ./scripts/docker/telemetry/deep_learning_container.py /usr/local/bin/deep_learning_container.py
+COPY ./scripts/docker/telemetry/bash_telemetry.sh.template /tmp/bash_telemetry.sh.template
 ARG FRAMEWORK
 ARG FRAMEWORK_VERSION
 ARG CONTAINER_TYPE
@@ -244,8 +244,8 @@ ENV PATH="/var/lang/bin:$PATH" \
   LAMBDA_RUNTIME_DIR=/var/runtime \
   LANG=en_US.UTF-8 \
   TZ=:/etc/localtime
-COPY ./scripts/telemetry/deep_learning_container.py /usr/local/bin/deep_learning_container.py
-COPY ./scripts/telemetry/bash_telemetry.sh.template /tmp/bash_telemetry.sh.template
+COPY ./scripts/docker/telemetry/deep_learning_container.py /usr/local/bin/deep_learning_container.py
+COPY ./scripts/docker/telemetry/bash_telemetry.sh.template /tmp/bash_telemetry.sh.template
 ARG FRAMEWORK
 
 ARG FRAMEWORK_VERSION
@@ -304,8 +304,8 @@ ENV PATH="/var/lang/bin:/usr/local/bin:$PATH" \
   LAMBDA_RUNTIME_DIR=/var/runtime \
   LANG=en_US.UTF-8 \
   TZ=:/etc/localtime
-COPY ./scripts/telemetry/deep_learning_container.py /usr/local/bin/deep_learning_container.py
-COPY ./scripts/telemetry/bash_telemetry.sh.template /tmp/bash_telemetry.sh.template
+COPY ./scripts/docker/telemetry/deep_learning_container.py /usr/local/bin/deep_learning_container.py
+COPY ./scripts/docker/telemetry/bash_telemetry.sh.template /tmp/bash_telemetry.sh.template
 ARG FRAMEWORK
 ARG FRAMEWORK_VERSION
 ARG CONTAINER_TYPE
@@ -330,7 +330,7 @@ ENTRYPOINT ["/lambda_entrypoint.sh", "python", "-m", "awslambdaric"]
 CMD ["handler.handler"]
 
 # ============================================================
-# RIC preview builder (compile thread-mode RIC with C extensions)
+# RIC builder (compile multi-mode concurrency RIC with C extensions)
 # ============================================================
 FROM builder-base as builder-ric-preview
 ARG PYTHON_VERSION
@@ -349,31 +349,46 @@ RUN uv pip install --python /var/lang/bin/python${PYTHON_VERSION} --no-deps /tmp
   && rm /tmp/awslambdaric-${AWSLAMBDARIC_VERSION}.tar.gz
 
 # ============================================================
-# Preview: Base Python 3.13 (+ thread-mode RIC)
+# Base Python 3.13 (+ multi-mode concurrency RIC)
 # ============================================================
 FROM base-py3 as base-preview-py3
 ARG PYTHON_VERSION
 RUN rm -rf /var/lang/lib/python${PYTHON_VERSION}/site-packages/awslambdaric* \
   && rm -f /var/lang/lib/python${PYTHON_VERSION}/site-packages/runtime_client*.so
 COPY --from=builder-ric-preview /var/lang/lib/python${PYTHON_VERSION}/site-packages/ /var/lang/lib/python${PYTHON_VERSION}/site-packages/
-ENV AWS_LAMBDA_CONCURRENCY_MODE=thread
+# Default: process mode (RIC-native). Customers override per function at runtime
+# to thread (one shared model copy in GPU VRAM) or hybrid; function-config env
+# vars take precedence over this image default. In process/hybrid, VRAM scales
+# with worker count (one model copy + CUDA context per process) — size function
+# concurrency to the model's VRAM footprint.
+ENV AWS_LAMBDA_CONCURRENCY_MODE=process
 
 # ============================================================
-# Preview: CuPy Python 3.13 (+ thread-mode RIC)
+# CuPy Python 3.13 (+ multi-mode concurrency RIC)
 # ============================================================
 FROM cupy-py3 as cupy-preview-py3
 ARG PYTHON_VERSION
 RUN rm -rf /var/lang/lib/python${PYTHON_VERSION}/site-packages/awslambdaric* \
   && rm -f /var/lang/lib/python${PYTHON_VERSION}/site-packages/runtime_client*.so
 COPY --from=builder-ric-preview /var/lang/lib/python${PYTHON_VERSION}/site-packages/ /var/lang/lib/python${PYTHON_VERSION}/site-packages/
-ENV AWS_LAMBDA_CONCURRENCY_MODE=thread
+# Default: process mode (RIC-native). Customers override per function at runtime
+# to thread (one shared model copy in GPU VRAM) or hybrid; function-config env
+# vars take precedence over this image default. In process/hybrid, VRAM scales
+# with worker count (one model copy + CUDA context per process) — size function
+# concurrency to the model's VRAM footprint.
+ENV AWS_LAMBDA_CONCURRENCY_MODE=process
 
 # ============================================================
-# Preview: PyTorch Python 3.13 (+ thread-mode RIC)
+# PyTorch Python 3.13 (+ multi-mode concurrency RIC)
 # ============================================================
 FROM pytorch-py3 as pytorch-preview-py3
 ARG PYTHON_VERSION
 RUN rm -rf /var/lang/lib/python${PYTHON_VERSION}/site-packages/awslambdaric* \
   && rm -f /var/lang/lib/python${PYTHON_VERSION}/site-packages/runtime_client*.so
 COPY --from=builder-ric-preview /var/lang/lib/python${PYTHON_VERSION}/site-packages/ /var/lang/lib/python${PYTHON_VERSION}/site-packages/
-ENV AWS_LAMBDA_CONCURRENCY_MODE=thread
+# Default: process mode (RIC-native). Customers override per function at runtime
+# to thread (one shared model copy in GPU VRAM) or hybrid; function-config env
+# vars take precedence over this image default. In process/hybrid, VRAM scales
+# with worker count (one model copy + CUDA context per process) — size function
+# concurrency to the model's VRAM footprint.
+ENV AWS_LAMBDA_CONCURRENCY_MODE=process
diff --git a/docker/lambda/pytorch/pyproject.toml b/docker/lambda/pytorch/pyproject.toml
@@ -7,11 +7,13 @@ dependencies = [
     "av==16.1.0",
     "awslambdaric==4.0.0",
     "boto3==1.40.4",
+    "click==8.3.3",
     "diffusers==0.38.0",
     "librosa==0.11.0",
+    "msgpack==1.2.1",
     "numpy==2.4.2",
     "opencv-python-headless==4.13.0.92",
-    "pillow==12.2.0",
+    "pillow==12.3.0",
     "pip==26.0.1",
     "pip-licenses==5.5.1",
     "requests==2.33.0",
@@ -24,7 +26,7 @@ dependencies = [
     "torch==2.10.0",
     "torchaudio==2.10.0",
     "torchvision==0.25.0",
-    "transformers==5.2.0",
+    "transformers==5.5.0",
     "urllib3==2.7.0",
     "pytest==8.4.1",
 ]