add execution time

.github/workflows/publish.yml

@@ -22,13 +22,9 @@ jobs:
           matrix:
             include:
               - docker_file: Dockerfile
-                tag-suffix: cpu
-                latest-tag: latest
                 label: "latest"
               - docker_file: Dockerfile.cuda
-                tag-suffix: cuda
-                latest-tag: latest-cuda
-                label: "cuda"
+                label: "latest-cuda"
         steps:
             - uses: actions/checkout@v4
 
@@ -62,6 +58,7 @@ jobs:
                 push: ${{ github.event_name != 'pull_request' }}
                 tags: |
                   ${{ steps.meta.outputs.tags }}
-                  ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}:${{ matrix.latest-tag }}
+                  ${{ matrix.docker_file == 'Dockerfile' && matrix.label == 'latest' && env.REGISTRY }}/{{ env.IMAGE_NAME }}:latest
+                  ${{ matrix.docker_file == 'Dockerfile.cuda' && matrix.label == 'latest-cuda' && env.REGISTRY }}/{{ env.IMAGE_NAME }}:latest-cuda
                 labels: ${{ steps.meta.outputs.labels }}
                 platforms: linux/amd64,linux/arm64

src/routes/api/summarization.py

@@ -1,6 +1,7 @@
 from fastapi import APIRouter, Query, Depends
 from src.middleware.auth.auth import get_api_key
 from src.shared.shared import summarization_model
+import time
 
 router = APIRouter()
 
@@ -10,10 +11,14 @@ async def summarization(
     text: str,
     model_name: str = Query(None),
 ):
+    start_time = time.time()
     summarizer = summarization_model(model_name)
     try:
         summary = summarizer(text)
-        return {"res": summary}
+        return {
+            "execution_time": time.time() - start_time,
+            "res": summary,
+        }
 
     except Exception as e:
         print("Something went wrong: ", e)

src/routes/api/translation.py

@@ -1,6 +1,7 @@
 from fastapi import APIRouter, Query, Depends
 from src.middleware.auth.auth import get_api_key
 from src.shared.shared import translation_model
+import time
 
 router = APIRouter()
 
@@ -12,11 +13,15 @@ async def translation(
     input_language: str = Query("en", description="Input language"),
     output_language: str = Query("de", description="Output language"),
 ):
+    start_time = time.time()
     translator = translation_model(model_name, input_language, output_language)
 
     try:
         translation = translator(text)
-        return {"res": translation}
+        return {
+            "execution_time": time.time() - start_time,
+            "res": translation,
+        }
 
     except Exception as e:
         print("Something went wrong: ", e)