Use whisper's medium model (#40)

lehigh-university-libraries · Sep 19, 2024 · 8e6d7c8 · 8e6d7c8
1 parent 10475e8
commit 8e6d7c8
Show file tree

Hide file tree

Showing 3 changed files with 6 additions and 5 deletions.
diff --git a/ci/k8s/whisper.yaml b/ci/k8s/whisper.yaml
@@ -33,8 +33,7 @@ spec:
               memory: "256Mi"
               cpu: "500m"
             limits:
-              memory: "1Gi"
-              nvidia.com/gpu: 1
+              memory: "5Gi"
           ports:
             - containerPort: 8080
           readinessProbe:

diff --git a/examples/whisper/Dockerfile b/examples/whisper/Dockerfile
@@ -3,8 +3,6 @@ ARG DOCKER_REPOSITORY=local
 FROM ${DOCKER_REPOSITORY}/scyllaridae:${TAG} AS scyllaridae
 FROM ghcr.io/ggerganov/whisper.cpp:main-cuda
 
-RUN bash ./models/download-ggml-model.sh base.en
-
 COPY --from=scyllaridae /app /app
 COPY scyllaridae.yml /app
 COPY cmd.sh /app

diff --git a/examples/whisper/cmd.sh b/examples/whisper/cmd.sh
@@ -4,12 +4,16 @@
 
 set -eou pipefail
 
+if [ ! -f /app/models/ggml-medium.en.bin ]; then
+  bash ./models/download-ggml-model.sh medium.en
+fi
+
 input_temp=$(mktemp /tmp/whisper-input-XXXXXX)
 
 cat > "$input_temp"
 
 /app/main \
-  -m /app/models/ggml-base.en.bin \
+  -m /app/models/ggml-medium.en.bin \
   --output-vtt \
   -f "$input_temp" \
   --output-file "$input_temp" > /dev/null 2>&1