Running DeepSeek-V4-Flash-0731 (155 GB MoE) on a DGX Spark with vLLM-Moet 2-bit quantization - AI's narrative