4 pointsby jbarti3 hours ago1 comment
  • jbarti3 hours ago
    While learning about different types of large language models I stumbled upon Bidirectional Encoder's (so called BERT models).

    Before, I used to always think you needed a ton of computing resources to do anything LLM related locally, but these models fascinated me because they could effectively run using only a few GBs of RAM, and without any GPU acceleration.

    As an experimental project I wanted to see how efficient it would be at analyzing Reddit threads and ended up with a pretty cool tool.

    This published version uses the `j-hartmann/emotion-english-distilroberta-base` fine tuned version of distiloberta-base.