生产中使用API问题
求教生产用大规模用bge-m3的话，这个用本身的哪个方法进行向量化呢，例如单个和批量向量化一个文本或多个文本呢？demo中的例子向量化方法感觉不大好用吧
The encode function of BGEM3FlagModel supports batch inference on multiple GPUs. To further accelerate the inference, you can use the TEI tool: https://github.com/huggingface/text-embeddings-inference .
The
encodefunction ofBGEM3FlagModelsupports batch inference on multiple GPUs. To further accelerate the inference, you can use the TEI tool: https://github.com/huggingface/text-embeddings-inference .
where specifically?
The
encodefunction ofBGEM3FlagModelsupports batch inference on multiple GPUs. To further accelerate the inference, you can use the TEI tool: https://github.com/huggingface/text-embeddings-inference .where, specifically?
If the bottleneck is the embedding API itself, I open-sourced m3serve a lightweight BGE-M3 server with batching https://github.com/MauroCE/m3serve as a simple start