JavaScript has many high level frameworks to make developers' lives easier by simplifying complex tasks. While it's often quick to set things up, frameworks can hide a lot of complexity under the hood. This can be a challenge when you need to debug or dive deep to customize something.

AI / ML is following a similar path. Incredible tools like Hugging Face Transformers, PEFT, LoRA, BitsAndBytes, and more have made training transformer models much easier by abstracting complex implementation details. But, as with JavaScript frameworks, this abstraction can sometimes leave you disconnected from what's happening behind the scenes.

Take Hugging Face's Trainer function— with over 100 parameters, it gives you an idea of just how much is going on in the background. 🫨

So, how do you balance using high-level frameworks and doing things manually? What’s your approach?

#AI #HuggingFace #Developers #Programming