Sentiment Analysis- The Emotional aspect of Social Media, Website Reviews, and Games- Part 8
Primer
As I discussed before, sentiment analysis is used to read emotions expressed as a text or voice data, using many techniques which include NLP, semantic search engines, dynamic classification, translation etc., which in turn takes us towards the opinion mining architecture, which I introduced in Part -1 of this series. The image below is to kindle your memory, and I will continue at GPT scaling, as I mentioned in my previous article.
Scaling in LLMs
The scaling in large language models such as GPT or BARD or BERT has not always been easy as fine tuning is not possible, hence results in poor predictability and large losses. Hence optimization techniques are usually applied to overcome the performance issues with scaling, and this is done using a scaling law with an irreducible loss term. In this type of optimization, the loss neither increases nor decreases, while scaled, thus having a constant performance during scaling. Check the image below for loss calculations while scaling.
