Update Dockerfile

Dockerfile

@@ -13,9 +13,13 @@ RUN unzip llama_bin.zip && rm llama_bin.zip && chmod +x bin/*
 
 ENV LD_LIBRARY_PATH=/app/bin:$LD_LIBRARY_PATH
 
-# Tải mô hình trong runtime thay vì build time
-RUN mkdir -p /models
-CMD ["sh", "-c", "wget -O /models/qwen2.5-0.5b-instruct-q5_k_m.gguf https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct-GGUF/resolve/main/qwen2.5-0.5b-instruct-q5_k_m.gguf && ./bin/llama-server -m /models/qwen2.5-0.5b-instruct-q5_k_m.gguf --host 0.0.0.0 --port 3000"]
+# Tạo thư mục và tải mô hình trong build time, đảm bảo quyền
+RUN mkdir -p /models && \
+    wget -O /models/qwen2.5-0.5b-instruct-q5_k_m.gguf https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct-GGUF/resolve/main/qwen2.5-0.5b-instruct-q5_k_m.gguf && \
+    chmod -R 777 /models
 
 # Dùng cổng 3000 cho Spaces
-EXPOSE 3000
+EXPOSE 3000
+
+# Chạy llama-server
+CMD ["./bin/llama-server", "-m", "/models/qwen2.5-0.5b-instruct-q5_k_m.gguf", "--host", "0.0.0.0", "--port", "3000"]