Fix NLTK punkt_tab by setting NLTK_DATA and pre-downloading

Dockerfile

@@ -4,33 +4,40 @@ ENV DEBIAN_FRONTEND=noninteractive PIP_NO_CACHE_DIR=1
 WORKDIR /app
 ENV PYTHONPATH=/app/backend:$PYTHONPATH
 
-# OpenMP runtime (prevents missing libgomp)
+# OpenMP runtime
 RUN apt-get update && apt-get install -y libgomp1 && rm -rf /var/lib/apt/lists/*
 
-# Writable caches + sane threads
+# Writable caches + thread caps + NLTK data path
 ENV HF_HOME=/app/.cache \
-    XDG_CACHE_HOME=/app/.cache \
+    TRANSFORMERS_CACHE=/app/.cache/transformers \
+    HUGGINGFACE_HUB_CACHE=/app/.cache/huggingface \
     SENTENCE_TRANSFORMERS_HOME=/app/.cache/sentence-transformers \
-    OMP_NUM_THREADS=1
-RUN mkdir -p /app/.cache && chmod -R 777 /app/.cache
+    XDG_CACHE_HOME=/app/.cache \
+    NLTK_DATA=/usr/local/nltk_data \
+    OMP_NUM_THREADS=1 \
+    OPENBLAS_NUM_THREADS=1 \
+    MKL_NUM_THREADS=1 \
+    NUMEXPR_NUM_THREADS=1
+
+RUN mkdir -p /app/.cache "$NLTK_DATA" && chmod -R 777 /app/.cache "$NLTK_DATA"
 
 COPY . .
 
-# Install GPU-ready deps
+# Install deps (CUDA 12.1 torch per requirements.txt)
 RUN pip install -r requirements.txt
 
-# Preload spaCy + NLTK data so runtime doesn't download
+# Preload spaCy + NLTK data into the path NLTK will search
 RUN python -m spacy download en_core_web_sm
 RUN python - <<'PY'
-import nltk
-nltk.download('punkt')
-nltk.download('punkt_tab')  # NEW: required by NLTK 3.9+
-nltk.download('wordnet')
-nltk.download('omw-1.4')
+import os, nltk
+path = os.environ.get("NLTK_DATA", "/usr/local/nltk_data")
+for pkg in ["punkt", "punkt_tab", "wordnet", "omw-1.4"]:
+    nltk.download(pkg, download_dir=path)
+print("Downloaded NLTK data to:", path)
 PY
 
-# HF Spaces uses port 7860
+# HF Spaces port
 EXPOSE 7860
 
-# Single worker + a few threads = nicer on GPU VRAM
+# Single worker + a few threads
 CMD ["gunicorn","-w","1","-k","gthread","--threads","4","-b","0.0.0.0:7860","backend.app:app"]