[Perf] Batched projector for pooling model embed, 1.8% throughput improvement#39533

yewentao256 wants to merge 6 commits intomainvllm-project/vllm:mainfrom

wentao-optimize-pooling-by-ragged-tensorvllm-project/vllm:wentao-optimize-pooling-by-ragged-tensorCopy head branch name to clipboard