部署BGE-VL-MLLM-S1主要分为四个步骤：
- environmental preparation: 创建Python3.10虚拟环境，安装transformers==4.41.2和sentencepiece
- Model loading:
from transformers import AutoModel, AutoProcessor
model = AutoModel.from_pretrained('BAAI/BGE-VL-MLLM-S1', trust_remote_code=True)
processor = AutoProcessor.from_pretrained(model_name, trust_remote_code=True) - input processing: 使用processor同时处理查询图像和文本，注意设置q_or_c参数区分查询/候选数据
- Similarity calculation:
query_embs = model(**query_inputs).hidden_states[-1][:, -1, :]
scores = torch.matmul(query_embs, candi_embs.T)
关键细节：必须使用GPU加速（需16GB+显存），处理文本时要包含任务指令模板，相似度计算前需对向量做L2归一化。
This answer comes from the articleMegaPairs: a new multimodal vector embedding model from BGEThe
May not be reproduced without permission:AI productivity tools " 如何快速部署BGE-VL-MLLM-S1模型进行图像检索？
English 
简体中文 
日本語 
Deutsch 
Português do Brasil