Disable HF fast downloader (or install hf_transfer) and make model prefetch non-fatal

Dockerfile

@@ -1,20 +1,17 @@
 FROM python:3.12-slim
 
-ENV DEBIAN_FRONTEND=noninteractive \
-    PIP_NO_CACHE_DIR=1 \
-    PYTHONUNBUFFERED=1
+ENV DEBIAN_FRONTEND=noninteractive PIP_NO_CACHE_DIR=1
 
-# Non-root user so caches are writable at runtime
+# Create non-root user
 RUN useradd -m -u 1000 appuser
 ENV PATH="/home/appuser/.local/bin:$PATH"
 
 WORKDIR /app
 
-# Minimal OS deps (OpenMP runtime)
-RUN apt-get update && apt-get install -y --no-install-recommends libgomp1 \
- && rm -rf /var/lib/apt/lists/*
+# Minimal OS deps
+RUN apt-get update && apt-get install -y --no-install-recommends libgomp1 && rm -rf /var/lib/apt/lists/*
 
-# Put ALL caches in the user's home; plus perf/env knobs
+# Put caches in user's home + sane threading + pythonpath
 ENV HF_HOME=/home/appuser/.cache \
     TRANSFORMERS_CACHE=/home/appuser/.cache/transformers \
     HUGGINGFACE_HUB_CACHE=/home/appuser/.cache/huggingface \
@@ -24,19 +21,19 @@ ENV HF_HOME=/home/appuser/.cache \
     OMP_NUM_THREADS=1 \
     OPENBLAS_NUM_THREADS=1 \
     MKL_NUM_THREADS=1 \
-    HF_HUB_ENABLE_HF_TRANSFER=1 \
-    PYTHONPATH=/app/backend:$PYTHONPATH
+    PYTHONPATH=/app/backend:$PYTHONPATH \
+    HF_HUB_ENABLE_HF_TRANSFER=0   # disable fast downloader to avoid missing package error
 
-# Copy code with correct ownership
+# Copy code as appuser
 COPY --chown=appuser:appuser . .
 
-# Switch to non-root BEFORE installs so caches land in /home/appuser
+# Switch to non-root BEFORE installs so caches/dirs are writable
 USER appuser
 
-# Python deps
+# Python deps (user site)
 RUN pip install --no-cache-dir --user -r requirements.txt
 
-# Preload NLP data
+# Preload spaCy + NLTK data
 RUN python -m spacy download en_core_web_sm
 RUN python - <<'PY'
 import nltk
@@ -45,19 +42,20 @@ for pkg in ["punkt","punkt_tab","wordnet","omw-1.4"]:
 print("NLTK OK")
 PY
 
-# Pre-fetch models to avoid cold-start timeouts
+# Prefetch models (don't fail the build if network hiccups)
 RUN python - <<'PY'
 from sentence_transformers import SentenceTransformer
 from transformers import AutoTokenizer, AutoModelForTokenClassification
-# semantic encoder
-SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
-# punctuation model
-AutoTokenizer.from_pretrained("oliverguhr/fullstop-punctuation-multilang-large")
-AutoModelForTokenClassification.from_pretrained("oliverguhr/fullstop-punctuation-multilang-large")
-print("HF models cached")
+try:
+    SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
+    AutoTokenizer.from_pretrained("oliverguhr/fullstop-punctuation-multilang-large")
+    AutoModelForTokenClassification.from_pretrained("oliverguhr/fullstop-punctuation-multilang-large")
+    print("HF models cached")
+except Exception as e:
+    print("Prefetch skipped:", e)
 PY
 
 EXPOSE 7860
 
-# Give the worker time for initial GPU warmup
+# Give cold start more time
 CMD ["gunicorn","-w","1","-k","gthread","--threads","4","--timeout","300","-b","0.0.0.0:7860","backend.app:app"]