Remove both Google Vertex & AWS SageMaker from Dockerfile-cuda-all

Neither of those is required, since both Google Cloud and AWS SageMaker
port the `Dockerfile-cuda-all` as-is, and then re-builds it there,
meaning that the actual BUILD_ARG for VERTEX is not being used at all,
neither the AWS SageMaker stage. For more information check the
repositories
https://github.com/huggingface/Google-Cloud-Containers/tree/main/containers/tei,
and
https://github.com/awslabs/llm-hosting-container/tree/main/huggingface/pytorch/tei/docker,
respectively for Google Cloud and AWS SageMaker.

Author: alvarobartt

Dockerfile-cuda-all

@@ -35,7 +35,6 @@ FROM base-builder AS builder
 
 ARG GIT_SHA
 ARG DOCKER_LABEL
-ARG VERTEX="false"
 
 # sccache specific variables
 ARG SCCACHE_GHA_ENABLED
@@ -51,39 +50,19 @@ COPY --from=planner /usr/src/recipe.json recipe.json
 
 RUN --mount=type=secret,id=actions_results_url,env=ACTIONS_RESULTS_URL \
     --mount=type=secret,id=actions_runtime_token,env=ACTIONS_RUNTIME_TOKEN \
-    if [ $VERTEX = "true" ]; \
-    then \
-      cargo chef cook --release --features google --recipe-path recipe.json && sccache -s; \
-    else \
-      cargo chef cook --release --recipe-path recipe.json && sccache -s; \
-    fi;
+    cargo chef cook --release --recipe-path recipe.json && sccache -s;
 
 RUN --mount=type=secret,id=actions_results_url,env=ACTIONS_RESULTS_URL \
     --mount=type=secret,id=actions_runtime_token,env=ACTIONS_RUNTIME_TOKEN \
-    if [ $VERTEX = "true" ]; \
-    then \
-      CUDA_COMPUTE_CAP=75 cargo chef cook --release --features google --features candle-cuda-turing --recipe-path recipe.json && sccache -s; \
-    else \
-      CUDA_COMPUTE_CAP=75 cargo chef cook --release --features candle-cuda-turing --recipe-path recipe.json && sccache -s; \
-    fi;
+    CUDA_COMPUTE_CAP=75 cargo chef cook --release --features candle-cuda-turing --recipe-path recipe.json && sccache -s;
 
 RUN --mount=type=secret,id=actions_results_url,env=ACTIONS_RESULTS_URL \
     --mount=type=secret,id=actions_runtime_token,env=ACTIONS_RUNTIME_TOKEN \
-    if [ $VERTEX = "true" ]; \
-    then \
-      CUDA_COMPUTE_CAP=80 cargo chef cook --release --features google --features candle-cuda --recipe-path recipe.json && sccache -s; \
-    else \
-      CUDA_COMPUTE_CAP=80 cargo chef cook --release --features candle-cuda --recipe-path recipe.json && sccache -s; \
-    fi;
+    CUDA_COMPUTE_CAP=80 cargo chef cook --release --features candle-cuda --recipe-path recipe.json && sccache -s;
 
 RUN --mount=type=secret,id=actions_results_url,env=ACTIONS_RESULTS_URL \
     --mount=type=secret,id=actions_runtime_token,env=ACTIONS_RUNTIME_TOKEN \
-    if [ $VERTEX = "true" ]; \
-    then \
-      CUDA_COMPUTE_CAP=90 cargo chef cook --release --features google --features candle-cuda --recipe-path recipe.json && sccache -s; \
-    else \
-      CUDA_COMPUTE_CAP=90 cargo chef cook --release --features candle-cuda --recipe-path recipe.json && sccache -s; \
-    fi;
+    CUDA_COMPUTE_CAP=90 cargo chef cook --release --features candle-cuda --recipe-path recipe.json && sccache -s;
 
 COPY backends backends
 COPY core core
@@ -93,34 +72,19 @@ COPY Cargo.lock ./
 
 RUN --mount=type=secret,id=actions_results_url,env=ACTIONS_RESULTS_URL \
     --mount=type=secret,id=actions_runtime_token,env=ACTIONS_RUNTIME_TOKEN \
-    if [ $VERTEX = "true" ]; \
-    then \
-        CUDA_COMPUTE_CAP=75 cargo build --release --bin text-embeddings-router -F candle-cuda-turing -F google  && sccache -s; \
-    else \
-        CUDA_COMPUTE_CAP=75 cargo build --release --bin text-embeddings-router -F candle-cuda-turing && sccache -s; \
-    fi;
+    CUDA_COMPUTE_CAP=75 cargo build --release --bin text-embeddings-router -F candle-cuda-turing && sccache -s;
 
 RUN mv /usr/src/target/release/text-embeddings-router /usr/src/target/release/text-embeddings-router-75
 
 RUN --mount=type=secret,id=actions_results_url,env=ACTIONS_RESULTS_URL \
     --mount=type=secret,id=actions_runtime_token,env=ACTIONS_RUNTIME_TOKEN \
-    if [ $VERTEX = "true" ]; \
-    then \
-        CUDA_COMPUTE_CAP=80 cargo build --release --bin text-embeddings-router -F candle-cuda -F google  && sccache -s; \
-    else \
-        CUDA_COMPUTE_CAP=80 cargo build --release --bin text-embeddings-router -F candle-cuda && sccache -s; \
-    fi;
+    CUDA_COMPUTE_CAP=80 cargo build --release --bin text-embeddings-router -F candle-cuda && sccache -s;
 
 RUN mv /usr/src/target/release/text-embeddings-router /usr/src/target/release/text-embeddings-router-80
 
 RUN --mount=type=secret,id=actions_results_url,env=ACTIONS_RESULTS_URL \
     --mount=type=secret,id=actions_runtime_token,env=ACTIONS_RUNTIME_TOKEN \
-    if [ $VERTEX = "true" ]; \
-    then \
-        CUDA_COMPUTE_CAP=90 cargo build --release --bin text-embeddings-router -F candle-cuda -F google  && sccache -s; \
-    else \
-        CUDA_COMPUTE_CAP=90 cargo build --release --bin text-embeddings-router -F candle-cuda && sccache -s; \
-    fi;
+    CUDA_COMPUTE_CAP=90 cargo build --release --bin text-embeddings-router -F candle-cuda && sccache -s;
 
 RUN mv /usr/src/target/release/text-embeddings-router /usr/src/target/release/text-embeddings-router-90
 
@@ -142,16 +106,6 @@ COPY --from=builder /usr/src/target/release/text-embeddings-router-75 /usr/local
 COPY --from=builder /usr/src/target/release/text-embeddings-router-80 /usr/local/bin/text-embeddings-router-80
 COPY --from=builder /usr/src/target/release/text-embeddings-router-90 /usr/local/bin/text-embeddings-router-90
 
-# Amazon SageMaker compatible image
-FROM base AS sagemaker
-
-COPY --chmod=775 sagemaker-entrypoint-cuda-all.sh entrypoint.sh
-
-ENTRYPOINT ["./entrypoint.sh"]
-
-# Default image
-FROM base
-
 COPY --chmod=775 cuda-all-entrypoint.sh entrypoint.sh
 
 ENTRYPOINT ["./entrypoint.sh"]