Gemma 4 QAT on One TPU v5e: What Runs and What Doesn't

This article provides a step by step guide to repacking Google's quantization-aware-trained (QAT) Gemma 4 weights for vLLM and serving them on one Google Cloud TPU v5e chip, with every build scored for classification, math, tool calling, throughput and long prompts. Every per-record output, log and script is committed. On one v5e chip the repacked... Weiterlesen: Gemma 4 QAT on One TPU v5e: What Runs and What Doesn't