phantomgrapes / mgeo Goto Github PK

rt

A classification model has been developed using MGeo:

Foundation Model: mgeo_backbone_chinese_base
Number of Classes: 270,000
Trained Model Size: about 1.2 GB
GPU Employed: A800
Inference Performance: the response time for a single request is roughly 238ms.
Request: Seeking support for converting the MGeo model to ONNX format, or exploring alternative methods to enhance the model's inference speed.

I found one code mistake in prepare_data/calculate_geographic_context.py

The highlighted part should be ly2, instead of ly1.

1、MGeo论文中似乎测评了GeoGlue的两个任务，ranking对应query-poi相关性任务，retrieval对应query-poi库召回任务，但是在ModelScope的网站没有放出query-poi库召回任务的模型，请问这个模型应该如何获得，我是否可以直接使用'damo/mgeo_backbone_chinese_base'底座，结合Retrieval.py来复现得到论文的结果呢？
2、’'damo/mgeo_backbone_chinese_base'‘这个底座是基于全国数据还是杭州数据呢

batch response: This repository is over its data quota. Account responsible for LFS bandwidth should purchase more data packs to restore access.

The recall data GeoTES-recall-test.json.zip download from https://modelscope.cn/datasets/damo/GeoGLUE/files does not provide pos_id. Can you provide the ground truth for the test querys?

cd data
unzip datasets.zip