SimKGC
SimKGC copied to clipboard
intfloat
ACL 2022, SimKGC: Simple Contrastive Knowledge Graph Completion with Pre-trained Language Models
hi☺. This Tree diagram is from its classes and can be useful for better understanding.
Hi! The paper argues that in-batch samples can be used as negatives. In the implementation of training, it seems there's no explicit positives, just mm() for all hr_vectors and tail_vectors?
Hello, if you train directly according to your steps, which one of IB, PB, SN will it turn out? If I want to test the results of different negative samples,...
It seems that cosine similarity was not used during training.These are some of the logits I obtained during repro training, and they are significantly greater than 1. logits tensor([[15.2290, 7.5998,...
Could you please explain how you handle relations in the code for the Bert.Tokenizer? Is it simply by adding "inverse" in front of each relation? It seems like you don't...
Hello, one of the assignments in our BDA course requires us to reproduce a paper, we want to choose yours, but we all only have our own laptops, and seeing...
I have a question about code implementation. First, in the doc.py section ```python def vectorize(self) -> dict: head_desc, tail_desc = self.head_desc, self.tail_desc if args.use_link_graph: if len(head_desc.split()) < 20: head_desc +=...
Thank you for sharing the code! After we have trained a model, how to use this trained model to obtain entity embeddings? Should we use the following function: ```python def...
I encountered the following problem when executing "UTPUT_DIR=./checkpoint/fb15k237/ bash scripts/train_fb.sh".O How can I solve it TypeError: Caught TypeError in replica 0 on device 0. TypeError: forward() got an unexpected keyword...
I want to ask a very simple but important question for me, that is, as a novice, I can't accurately find the three negative sampling parts in the code. Can...
