Traversing the ML World Without a Map — Part 0
Well to start, I didn’t have much of a goal for this project. I just wanted to try and fine-tune — or train, as I didn’t even know what fine-tuning was back then. I had just watched Pewds’ video about how he was doing it, and surely I cannot do it on that level because I don’t have that much money just lying around.
But I did get a laptop with a GPU not so long ago, and I knew I could do something with it. So here is my journey navigating the Machine Learning world without a map — essentially.
To start, I just have a Lenovo LOQ gaming laptop which has an RTX 3050 with 6GB VRAM, which does work for lower end models. At first what I thought was to make an assistant — something that could run locally. What a fool I was back then.
So as anyone would do, I went straight to Claude to ask about how I would go about this project, and realized this was something that requires way more hardware than I currently have. But I still wanted to do it, so I went through the ways of doing it and Claude told me I could take an open source model and feed it data during runtime and get results out of it.
This was a start. I then made a list of models I could run on my GPU, out of which I decided to go with Qwen3 1.7B base model — suggested by Claude.
Now everything was almost ready for me to start.
This is where I end Part 0 — stay tuned for more revelations about machine learning, LLMs, and also where everything just does not work for some reason.