Fantastic Tutorial mate you are a natural teacher. I was clicking around Huggingface like a noob for hours until I found your video. Thanks so much for this.
Hey, glad you enjoyed it. Sure, are you after a video on using RoBERTa on 'The Microsoft Research Sentence Completion Challenge' ? Happy to try get that done soon as a next video :)
Hi Rupert, I liked your video. I have one question, sorry if this is something you already answered, I want to create my own bert model from scratch using a specific dataset that's based on my use case. Any tutorials could you please suggest that I can follow for that? Thank you so much in advance.
you just need to stack some transformer encoders , and train them using your dataset with two specified tasks in the original bert paper, namely NSP and MLM.
Hi Karan, sure - it sort of depends what dataset you are using and what task you are trying to achieve? If you look at my other video on multi-label classification then you can get a good template for doing any task with hugging face with any dataset! Let me know if I can help any further :)
you dont need to build bert from scratch if you're only looking to re-train with your own dataset. You can add a machine learning layer on top of BERT where bert is used for word embedding and your layer(s) is used for optimizing BERT based on a data set you provide.
bro ,am trying to create a speech to text system where only like 30 words need to be recognized...from what i know CNN can do the trick..but do i need to go for end to end models for speech recognition.hope you could reply me asap
Great video! Is it possible to apply either the manual or the pre-made pipeline approach to a dataframe/dataset, instead of just those small lists? I wanted to solve a binary classification task but have 2 different text columns. Thanks for the video, good stuff man!