inference speed

gte-multilingual-reranker-base inference speed is slow, and it is not as fast as using transformers directly.

Start by reproducing the reported gte-multilingual-reranker-base inference-speed difference and record the model, environment, request shape, and comparison with direct Transformers usage. The issue names no files, tests, or entry points, so define a reproducible benchmark and identify the relevant inference path before determining what improvement would count as done.

Written by the indexing model from the issue text.