feat: add all methods to vertex API (#192)

huggingface · Mar 21, 2024 · a57cf61 · a57cf61
1 parent 5e60d06
commit a57cf61
Show file tree

Hide file tree

Showing 6 changed files with 505 additions and 122 deletions.
diff --git a/Dockerfile-cuda-all b/Dockerfile-cuda-all
@@ -33,6 +33,7 @@ FROM base-builder AS builder
 
 ARG GIT_SHA
 ARG DOCKER_LABEL
+ARG VERTEX
 
 # sccache specific variables
 ARG ACTIONS_CACHE_URL
@@ -45,39 +46,69 @@ COPY --from=planner /usr/src/recipe.json recipe.json
 
 FROM builder as builder-75
 
-RUN CUDA_COMPUTE_CAP=75 cargo chef cook --release --features candle-cuda-turing --no-default-features --recipe-path recipe.json && sccache -s
+RUN if [ $VERTEX = "true" ]; \
+    then \
+      CUDA_COMPUTE_CAP=75 cargo chef cook --release --features google --features candle-cuda-turing --no-default-features --recipe-path recipe.json && sccache -s; \
+    else \
+      CUDA_COMPUTE_CAP=75 cargo chef cook --release --features candle-cuda-turing --no-default-features --recipe-path recipe.json && sccache -s; \
+    fi;
 
 COPY backends backends
 COPY core core
 COPY router router
 COPY Cargo.toml ./
 COPY Cargo.lock ./
 
-RUN CUDA_COMPUTE_CAP=75 cargo build --release --bin text-embeddings-router -F candle-cuda-turing -F http --no-default-features && sccache -s
+RUN if [ $VERTEX = "true" ]; \
+    then \
+        CUDA_COMPUTE_CAP=75 cargo build --release --bin text-embeddings-router -F candle-cuda-turing -F http -F google --no-default-features && sccache -s; \
+    else \
+        CUDA_COMPUTE_CAP=75 cargo build --release --bin text-embeddings-router -F candle-cuda-turing -F http --no-default-features && sccache -s; \
+    fi;
 
 FROM builder as builder-80
 
-RUN CUDA_COMPUTE_CAP=80 cargo chef cook --release --features candle-cuda --no-default-features --recipe-path recipe.json && sccache -s
+RUN if [ $VERTEX = "true" ]; \
+    then \
+      CUDA_COMPUTE_CAP=80 cargo chef cook --release --features google --features candle-cuda-turing --no-default-features --recipe-path recipe.json && sccache -s; \
+    else \
+      CUDA_COMPUTE_CAP=80 cargo chef cook --release --features candle-cuda-turing --no-default-features --recipe-path recipe.json && sccache -s; \
+    fi;
 
 COPY backends backends
 COPY core core
 COPY router router
 COPY Cargo.toml ./
 COPY Cargo.lock ./
 
-RUN CUDA_COMPUTE_CAP=80 cargo build --release --bin text-embeddings-router -F candle-cuda -F http --no-default-features && sccache -s
+RUN if [ $VERTEX = "true" ]; \
+    then \
+        CUDA_COMPUTE_CAP=80 cargo build --release --bin text-embeddings-router -F candle-cuda-turing -F http -F google --no-default-features && sccache -s; \
+    else \
+        CUDA_COMPUTE_CAP=80 cargo build --release --bin text-embeddings-router -F candle-cuda-turing -F http --no-default-features && sccache -s; \
+    fi;
 
 FROM builder as builder-90
 
-RUN CUDA_COMPUTE_CAP=90 cargo chef cook --release --features candle-cuda --no-default-features --recipe-path recipe.json && sccache -s
+RUN if [ $VERTEX = "true" ]; \
+    then \
+      CUDA_COMPUTE_CAP=90 cargo chef cook --release --features google --features candle-cuda-turing --no-default-features --recipe-path recipe.json && sccache -s; \
+    else \
+      CUDA_COMPUTE_CAP=90 cargo chef cook --release --features candle-cuda-turing --no-default-features --recipe-path recipe.json && sccache -s; \
+    fi;
 
 COPY backends backends
 COPY core core
 COPY router router
 COPY Cargo.toml ./
 COPY Cargo.lock ./
 
-RUN CUDA_COMPUTE_CAP=90 cargo build --release --bin text-embeddings-router -F candle-cuda -F http --no-default-features && sccache -s
+RUN if [ $VERTEX = "true" ]; \
+    then \
+        CUDA_COMPUTE_CAP=90 cargo build --release --bin text-embeddings-router -F candle-cuda-turing -F http -F google --no-default-features && sccache -s; \
+    else \
+        CUDA_COMPUTE_CAP=90 cargo build --release --bin text-embeddings-router -F candle-cuda-turing -F http --no-default-features && sccache -s; \
+    fi;
 
 FROM nvidia/cuda:12.2.0-runtime-ubuntu22.04 as base