Update Dockerfile

Dockerfile

@@ -3,11 +3,7 @@ FROM nvidia/cuda:12.5.1-cudnn-devel-ubuntu20.04
 
 # Set environment variables for non-interactive installation and timezone
 ENV DEBIAN_FRONTEND=noninteractive \
-    TZ=Europe/Paris \
-    VLLM_CPU_KVCACHE_SPACE=40 \
-    VLLM_CPU_OMP_THREADS_BIND=29 \
-    OMP_NUM_THREADS=4 \
-    MKL_NUM_THREADS=4
+    TZ=Europe/Paris
 
 # Remove third-party apt sources to avoid issues with expiring keys
 # Install basic utilities
@@ -90,9 +86,8 @@ RUN pip install --upgrade pip && \
 RUN pip install vllm && \
     pip install huggingface-hub[cli]
 
-
 # Expose port for the model server
-EXPOSE 7860
+EXPOSE 8007
 
 # Set the working directory to /app and set the model directory
 WORKDIR $HOME/app