Defer model downloads to runtime to support gated models via HF_TOKEN

Dockerfile

@@ -20,17 +20,6 @@ RUN pip install --no-cache-dir .
 RUN mkdir -p /app/nltk_data && \
     python -c "import nltk; nltk.download('punkt', download_dir='/app/nltk_data'); nltk.download('punkt_tab', download_dir='/app/nltk_data'); nltk.download('stopwords', download_dir='/app/nltk_data')"
 
-# Pre-download HF embedding models at build time (download files only, no RAM loading)
-RUN python -c "from huggingface_hub import snapshot_download; \
-    snapshot_download(repo_id='nomic-ai/nomic-embed-text-v1.5'); \
-    snapshot_download(repo_id='google/gemma-3n-embedding-exp'); \
-    snapshot_download(repo_id='Qwen/Qwen3-Embedding-0.6B')"
-
-# Pre-download HF Qwen LLM model at build time (download files only, no RAM loading)
-RUN python -c "from huggingface_hub import snapshot_download; \
-    snapshot_download(repo_id='Qwen/Qwen2.5-1.5B-Instruct')"
-
-
 # Copy app code
 COPY . .