Fix worker timeout and tokenizer fork issues

- Set TOKENIZERS_PARALLELISM=false to prevent fork deadlocks
- Increase gunicorn timeout to 600s for heavy model inference
- Use /dev/shm for worker heartbeat (avoids container filesystem issues)
- Single thread to minimize memory usage

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Dockerfile

@@ -16,7 +16,10 @@ RUN pip install --no-cache-dir -r requirements.txt
 COPY app.py .
 COPY index.html .
 
+# Avoid tokenizer fork deadlocks
+ENV TOKENIZERS_PARALLELISM=false
+
 # HuggingFace Spaces runs on port 7860
 EXPOSE 7860
 
-CMD ["gunicorn", "--bind", "0.0.0.0:7860", "--timeout", "300", "--workers", "1", "--preload", "--access-logfile", "-", "--error-logfile", "-", "app:app"]
+CMD ["gunicorn", "--bind", "0.0.0.0:7860", "--timeout", "600", "--workers", "1", "--threads", "1", "--preload", "--worker-tmp-dir", "/dev/shm", "--access-logfile", "-", "--error-logfile", "-", "app:app"]