Update Dockerfile

Dockerfile

@@ -1,57 +1,33 @@
-FROM python:3.12
-
-# Optional: zuerst llama-cpp-python bauen (cachebar)
-#RUN pip install --no-cache-dir --no-build-isolation llama-cpp-python
-# Danach: outetts (zieht llama-cpp-python nicht erneut)
-#RUN pip install --no-cache-dir --no-build-isolation outetts
+# GPU‑fertige Basis mit Python 3.10, CUDA 11.8, cuDNN 8
+FROM pytorch/pytorch:2.2.2-cuda11.8-cudnn8-runtime
 
+# System‑Tools (schlank halten!)
 RUN apt-get update && \
-    apt-get install -y \
-      bash \
-      git git-lfs \
-      wget curl procps gnupg \
-      build-essential cmake \
-      htop vim nano && \
+    apt-get install -y git-lfs build-essential && \
     rm -rf /var/lib/apt/lists/*
 
-# NVIDIA CUDA Keyring installieren (offizielle Methode seit 2024)
-RUN wget https://developer.download.nvidia.com/compute/cuda/repos/debian12/x86_64/cuda-keyring_1.1-1_all.deb && \
-    dpkg -i cuda-keyring_1.1-1_all.deb && \
-    apt-get update && \
-    apt-get -y install cuda
-
-# CUDA ENV-Variablen setzen
-ENV PATH=/usr/local/cuda/bin:${PATH}
-ENV LD_LIBRARY_PATH=/usr/local/cuda/lib64:${LD_LIBRARY_PATH}
-ENV CUDAToolkit_ROOT=/usr/local/cuda
-ENV CMAKE_ARGS="-DGGML_CUDA=on -DCMAKE_CUDA_ARCHITECTURES=86"
-
+# Non‑root‑User, weil Spaces das mögen
 RUN useradd -m -u 1000 user
 USER user
-ENV PATH="/home/user/.local/bin:$PATH"
-# ^ when run as `user`, pip installs executables there
-
 WORKDIR /app
+ENV PATH="/home/user/.local/bin:$PATH"
+ENV HF_HOME=/app/.cache  # HF‑Cache in deinem Schreib­verzeichnis
 
-COPY --chown=user . /app
-
-# Wichtig: Isolation deaktivieren für llama-cpp-python Build
-RUN pip install --upgrade pip
-
-# Manuell Build-Werkzeuge bereitstellen
+# Python-Abhängigkeiten
+COPY --chown=user requirements.txt .
 RUN pip install --upgrade pip && \
-    pip install \
-        setuptools \
-        wheel \
-        packaging \
-        ninja \
-        scikit-build-core[pyproject]
+    pip install --no-cache-dir -r requirements.txt
 
-#RUN pip install --no-cache-dir --no-build-isolation torch==2.2.2
+# Optional: flash‑attn (Ampere 86 wird erkannt, Wheel vorhanden)
+RUN pip install --no-cache-dir flash-attn==2.5.2 --no-build-isolation
 
-RUN pip install --no-cache-dir -r requirements.txt
+# Mount das geheime HF‑Token beim Build:
+#   Settings → Secrets → Name: HF_TOKEN  (scope: "read")
+#ARG HF_TOKEN
+#RUN --mount=type=secret,id=HF_TOKEN \
+#    echo "machine huggingface.co login __token__ password $(cat /run/secrets/HF_TOKEN)" > ~/.netrc
 
-RUN pip install flash-attn --no-build-isolation 
+# App‑Code
+COPY --chown=user . /app
 
-#CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860", , "--ws", "auto", "--allow-websocket-origin", "*"]
 CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]