Ask HN: Resource to learn how to train and use ML Models

15 points by ud0 3 years ago | 10 comments

I am an experienced software engineer with decent knowledge of Python. I've used Machine Learning Models in the past to do stuff like background blurring etc.

What do I want? I want to be able to train models and use them for the applications I intend to build. I honestly do not care about the Math, statistics or theory behind them. I just want to know enough to be able to train a model, use a model and productionize it. Most of the resources I have seen on the web are bloated with information that I consider irrelevant and don't care about. Are there any resources where I can simply learn to do these things?

tikkun 3 years ago |

I compiled this list - https://gist.github.com/TikkunCreation/5de1df7b24800cc05b482...

In particular, you'll probably want to skip to nanoGPT (https://github.com/karpathy/nanoGPT) and then maybe if you are interested in a bit more of the theory, Zero to Hero (https://karpathy.ai/zero-to-hero.html), and his comments in one of the threads linked: https://news.ycombinator.com/item?id=34414716

Fine tuning may also be a faster and better place to start, this is a good guide for fine tuning some publicly released LLMs: https://erichartford.com/uncensored-models

ud0 3 years ago | |

Thank you!

t14n 3 years ago |

https://fullstackdeeplearning.com/ and https://course.fast.ai/ seem to be what you're looking for.

Both only cover the math as it becomes relevant. I am working through fast.ai's book right now and find its pragmatic approach to DL pretty agreeable to just getting models hosted and out the door. I watched the lectures before hand, and there are several Jupyter notebooks and examples on how to get models deployed ASAP with clunky interfaces, which also might be of interest to you.

bazmattaz 3 years ago |

> I honestly do not care about the Math, statistics or theory behind them. I just want to know enough to be able to train a model, use a model and productionize it.

I have to say this comes across a little insulting to machine learning engineers. You’re asking for a quick snappy course that will teach what MLEs take years to learn and master. Sigh.

Nevertheless, The Andrew Ng course ML course on Coursera is a favourite among minds curious about ML.

Productionising a model is a whole different ball game and there is likely a wide range of content on the matter - it depends what you are trying to achieve, why does your live environment look like? How many users do you have? What SLOs and such if any do you have to meet?

ud0 3 years ago | |

> I have to say this comes across a little insulting to machine learning engineers.

How so? I have a lot of respect for ML engineers and I know what they do is no easy feat. However, I do not intend to get into ML, I see it as a tool to help me achieve an objective. My goal is to utilise said tool to achieve an objective, I have no intention of becoming an expert in this area. It's just like CSS when you want to use it, do you have to care about how the browser rendering works? No you don't, you simply use CSS.

thdespou 3 years ago | |

> I honestly do not care about the Math, statistics or theory behind them. I just want to know enough to be able to train a model, use a model and productionize it.

How can anyone do that without some understanding of the maths behind it? Just downloading some jupiter notebooks and running them on colab is considered good enough? You and other 100k people. I don't think so.

pontiacbandit8 3 years ago | |

!? How is that insulting to ML engineers?! When I want to just drive a formula one car do I need to understand the inner workings of a car engine? No.

How many C developers understand what a compiler does?! They just want the program to do what they want they don’t need to know how a code is the compiled.

I think your comment is a bit on a social warrior side.

segmondy 3 years ago |

Join the appropriate reddit groups machinelearning, locallamma, deeplearning, and follow folks in the ML space on twitter, plus use a search engine.

1. Learn to run a model, checkout llama.cpp Tons of free models on huggingface.com

2. Learn to finetune a model - https://github.com/lxe/simple-llm-finetuner

3. Learn to train one. PyTorch, TensorFlow, HuggingFace libraries, etc.

Good luck.

stillpoint 3 years ago |

https://madewithml.com/

jstx1 3 years ago |

The fastai courses are more on the pragmatic side. Not the biggest fan of their lecture order but many people like them.

The other option is to follow tutorials - e.g. pytorch.org/tutorials - it doesn’t get any more practical and light on theory than that.