Update Dockerfile

Dockerfile

@@ -24,9 +24,11 @@ RUN pip install --no-cache-dir --upgrade pip && \
       pillow \
       requests
 
-# Tạo stub module flash_attn để thoả mãn yêu cầu import của model mà không cần CUDA
+# Tạo stub module flash_attn và bitsandbytes để tránh lỗi trên môi trường CPU-only
 RUN mkdir -p /usr/local/lib/python3.10/site-packages/flash_attn && \
-    printf "def _not_available(*args, **kwargs):\n    raise ImportError('flash_attn is not available in this CPU-only build')\n\n__all__ = ['_not_available']\n" > /usr/local/lib/python3.10/site-packages/flash_attn/__init__.py
+    printf "def _not_available(*args, **kwargs):\n    raise ImportError('flash_attn is not available in this CPU-only build')\n\n__all__ = ['_not_available']\n" > /usr/local/lib/python3.10/site-packages/flash_attn/__init__.py && \
+    mkdir -p /usr/local/lib/python3.10/site-packages/bitsandbytes && \
+    printf \"raise ImportError('bitsandbytes is not available in this CPU-only build')\n\" > /usr/local/lib/python3.10/site-packages/bitsandbytes/__init__.py
 
 COPY app.py /app/app.py