Nils Reimers: Sentence Transformers, Search, Future of NLP | Learning from Machine Learning #3
Learning from Machine Learning
Train Sentence Encoder
The original sentence transformers model was just trained like small dataset from NLI. So that's what the first version of sentence transformers was trained on and sadly evaluated on. But now it was trained on like a whole bunch of text, like really ugly, noisy social media text full of hashtags and emojis. Now the model understands like emojis and hashtags and knows, okay, what's the similarity between hashtags and relationship in hashtags. Right. And this gives you a much, much better model,. sadly, sometimes if people use it on the old benchmarks on the nicely cleanly written text, doesn't perform as well as models overfitted on these settings.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
