User runs pruned DeepSeek-V4-Flash on Ascent GX10

Reddit user reports running a 162B REAP-pruned NVFP4 DeepSeek-V4-Flash on a single Ascent GX10 Spark, maintaining consistent long-context performance. The setup uses a patched spark-vllm-docker image with pruning by 0xSero.