Shamane Siri

Hi , I am using BERT based UDA. When I calculated the KL loss between augmented text and unaugmented text, it is similar to when I send the same text...

Did you use this loss ?

The forward function of the TransformerEncoderLayer can have **src_key_padding_mask**. Maybe we can update it too.

What kind of search method are you using? do you use faiss library?

What is the maximum input sequence length that the T5 model can handle?

1

Can we increase the sequence length?

Here, after a given number of episodes(Bath Size) we train the A3C agent with calculating the return. So we need to feed states, return, advantage function as a batch to...

DeepMind Retro

10

# 🌟 New model addition Basically a retrieval augmented model like RAG, but without expensive retriever end2end training ## Open source status * [ ] the model implementation is available:...