Update Dockerfile

Dockerfile

@@ -1,51 +1,52 @@
 FROM python:3.11-slim
 
 ENV PIP_NO_CACHE_DIR=1 \
-    PIP_DISABLE_PIP_VERSION_CHECK=1 \
     PYTHONDONTWRITEBYTECODE=1 \
     PYTHONUNBUFFERED=1 \
-    PYTHONOPTIMIZE=1 \
-    APP_CACHE=/app/.cache \
-    XDG_CACHE_HOME=/app/.cache \
-    HF_HOME=/app/.cache/hf \
-    HUGGINGFACE_HUB_CACHE=/app/.cache/hf \
-    TORCH_HOME=/app/.cache/torch \
-    OPENCLIP_CACHE_DIR=/app/.cache/open_clip \
+    APP_CACHE=/tmp/appcache \
+    XDG_CACHE_HOME=/tmp/appcache \
+    HF_HOME=/tmp/appcache/hf \
+    HUGGINGFACE_HUB_CACHE=/tmp/appcache/hf \
+    TORCH_HOME=/tmp/appcache/torch \
+    OPENCLIP_CACHE_DIR=/tmp/appcache/open_clip \
     HF_HUB_ENABLE_HF_TRANSFER=0 \
-    TOKENIZERS_PARALLELISM=false \
     OMP_NUM_THREADS=2 \
     MKL_NUM_THREADS=2 \
-    NUM_THREADS=2 \
-    OMP_PROC_BIND=true \
-    OMP_PLACES=cores
+    NUM_THREADS=2
 
-# deps del sistema
+# deps del sistema: OpenMP y build tools
 RUN apt-get update && apt-get install -y --no-install-recommends \
     build-essential git libgomp1 && \
     rm -rf /var/lib/apt/lists/*
 
 WORKDIR /app
 
-# preparar caches con permisos amplios
-RUN mkdir -p /app/.cache/hf /app/.cache/open_clip /app/.cache/torch && \
-    chmod -R 777 /app/.cache
-
-# deps de proyecto
+# deps de proyecto (ligeras)
 COPY requirements.txt ./
-RUN pip install --upgrade pip && \
-    pip install --no-cache-dir --prefer-binary -r requirements.txt
+RUN pip install --upgrade pip && pip install --no-cache-dir -r requirements.txt
 
-# pytorch cpu oficial
+# PyTorch CPU wheels oficiales
 RUN pip install --no-cache-dir --index-url https://download.pytorch.org/whl/cpu \
     torch torchvision
 
-# codigo y embeddings
+# (opcional) asegurar que open_clip este instalado (si no viene por reqs)
+# RUN pip install --no-cache-dir open_clip_torch
+
+# copiar codigo y embeddings
 COPY server1.py ./server1.py
 COPY text_embeddings_modelos_bigg.pt ./text_embeddings_modelos_bigg.pt
 COPY text_embeddings_bigg.pt ./text_embeddings_bigg.pt
 
+# (opcional) precarga ligera para evitar cold-start de dependencias
+# No hace forward pesado ni cambia precision
+# RUN python - <<'PY'\n\
+# import open_clip\n\
+# open_clip.create_model_and_transforms('hf-hub:timm/PE-Core-bigG-14-448', pretrained=None)\n\
+# print('open_clip model meta loaded')\n\
+# PY
+
 ENV PORT=7860
 EXPOSE 7860
 
-# uvicorn ligero; 1 worker para no duplicar el modelo
-CMD ["uvicorn", "server1:app", "--host", "0.0.0.0", "--port", "7860", "--loop", "uvloop", "--http", "httptools", "--workers", "1", "--no-server-header", "--no-access-log", "--log-level", "warning", "--timeout-keep-alive", "5"]
+# uvloop + httptools (vienen con uvicorn[standard]); 1 worker evita duplicar el modelo
+CMD ["uvicorn", "server1:app", "--host", "0.0.0.0", "--port", "7860", "--loop", "uvloop", "--http", "httptools", "--workers", "1", "--no-server-header", "--timeout-keep-alive", "5"]