27B vLLM on V100: Deployment and Benchmark Guide

Full Record: 4-GPU Tensor Parallelism, NVFP4, FP8 KV Cache & Speculative Decoding This article documents end-to-end deployment and stress-testing of a 27B multimodal large model on 8 Tesla V100-SXM2-3