FlagOpen / FlagEmbedding

Retrieval and Retrieval-augmented LLMs
MIT License
7.01k stars 512 forks source link

关于C-mteb评测数据 #342

Open zhaobinNF opened 9 months ago

zhaobinNF commented 9 months ago
image

您好,问下您还能评测这个mteb/amazon_reviews_multi数据吗,好像这个数据集已经disable了

staoxiao commented 9 months ago

mteb有一份自己的数据:https://huggingface.co/datasets/mteb/amazon_reviews_multi

zhaobinNF commented 9 months ago

{ "dataset_revision": null, "dev": { "evaluation_time": 1257.88, "map_at_1": 0.22166, "map_at_10": 0.32886, "map_at_100": 0.34724, "map_at_1000": 0.34865, "map_at_3": 0.2937, "map_at_5": 0.3128, "mrr_at_1": 0.34459, "mrr_at_10": 0.41874, "mrr_at_100": 0.42905, "mrr_at_1000": 0.42965, "mrr_at_3": 0.39602, "mrr_at_5": 0.40849, "ndcg_at_1": 0.34459, "ndcg_at_10": 0.38978, "ndcg_at_100": 0.46511, "ndcg_at_1000": 0.49128, "ndcg_at_3": 0.34527, "ndcg_at_5": 0.36272, "precision_at_1": 0.34459, "precision_at_10": 0.0874, "precision_at_100": 0.0149, "precision_at_1000": 0.00182, "precision_at_3": 0.19663, "precision_at_5": 0.14134, "recall_at_1": 0.22166, "recall_at_10": 0.48025, "recall_at_100": 0.79554, "recall_at_1000": 0.97433, "recall_at_3": 0.34388, "recall_at_5": 0.40053 }, "mteb_dataset_name": "CmedqaRetrieval", "mteb_version": "1.1.1" }我通过测试,得到了这样的结果,但是和没有找到与这个值对应的数据,请问应该比较哪个值呢

image
staoxiao commented 9 months ago

展示的是ndcg@10,如果测的是bge模型的话,需要加上指令, 参考脚本:https://github.com/FlagOpen/FlagEmbedding/tree/master/C_MTEB#evaluate-embedding-model