Fix model loading

Dockerfile

@@ -7,19 +7,18 @@ WORKDIR /app
 # Copy everything including smoldocling
 COPY . .
 
-# Ensure cache folders exist
-RUN mkdir -p /tmp/huggingface/transformers /tmp/huggingface/hub
+# Set cache environment so model is stored in /app/.cache
+ENV TRANSFORMERS_CACHE=/app/.cache/transformers
+ENV HF_HUB_CACHE=/app/.cache/hub
 
-# Set cache env vars
-ENV TRANSFORMERS_CACHE=/tmp/huggingface/transformers
-ENV HF_HUB_CACHE=/tmp/huggingface/hub
-
-# Install requirements (editable install of local package)
+# Pre-download the model
 RUN pip install --no-cache-dir -r requirements.txt && \
-    pip install -e ./smoldocling
+    pip install -e ./smoldocling && \
+    python -c "from transformers import AutoTokenizer, AutoModel; \
+    AutoTokenizer.from_pretrained('ds4sd/SmolDocling-256M-preview'); \
+    AutoModel.from_pretrained('ds4sd/SmolDocling-256M-preview')"
+
 
-# 🟩 Pre-download model to avoid runtime issues
-RUN python -c "from transformers import AutoModel, AutoTokenizer; AutoTokenizer.from_pretrained('ds4sd/SmolDocling-256M-preview'); AutoModel.from_pretrained('ds4sd/SmolDocling-256M-preview')"
 
 EXPOSE 7860
 

app.py

@@ -7,6 +7,9 @@ from smoldocling import cli
 import shutil
 import dotenv
 
+os.environ["TRANSFORMERS_CACHE"] = "/app/.cache/transformers"
+os.environ["HF_HUB_CACHE"] = "/app/.cache/hub"
+
 # Load environment variables
 dotenv.load_dotenv()