https://anandpatel.com/series/fine-tuning-at-home/ Training a small open model to write in a specific voice, from what a language model is made of through to whether the result actually worked. Eleven parts, illustrated, assuming no prior AI knowledge and a machine far smaller than mine. en-us Tue, 01 Sep 2026 12:10:00 GMT https://anandpatel.com/series/fine-tuning-at-home/01-how-a-language-model-works/ https://anandpatel.com/series/fine-tuning-at-home/01-how-a-language-model-works/ Tue, 01 Sep 2026 12:00:00 GMT The whole machine end to end, in one figure: text becomes tokens, tokens become vectors, a stack of identical blocks transforms them, and the top produces a probability for every possible next word. Start here if any of that is new. Everything after it is one stage of this part, enlarged. https://anandpatel.com/series/fine-tuning-at-home/02-matrix-multiplication/ https://anandpatel.com/series/fine-tuning-at-home/02-matrix-multiplication/ Tue, 01 Sep 2026 12:01:00 GMT A short refresher: how to read a shape like <code>[5, 4096]</code>, one worked example small enough to check by hand, and how to count a model's parameters yourself. Not a mathematics course and it says so. <b>Skip it</b> if you write software and remember your linear algebra. https://anandpatel.com/series/fine-tuning-at-home/03-tokens-and-embeddings/ https://anandpatel.com/series/fine-tuning-at-home/03-tokens-and-embeddings/ Tue, 01 Sep 2026 12:02:00 GMT The stage nobody thinks about, which explains more of the famous stupid failures than anything else in the architecture. Where the vocabulary came from, why rare words shatter into pieces, and the special tokens that a fine-tune gets wrong at its peril. https://anandpatel.com/series/fine-tuning-at-home/04-attention/ https://anandpatel.com/series/fine-tuning-at-home/04-attention/ Tue, 01 Sep 2026 12:03:00 GMT The only place in the entire architecture where one word can affect another, and therefore the mechanism behind everything context does. The part worth slowing down for. Also why long prompts cost what they do, and what that implies for how you write one. https://anandpatel.com/series/fine-tuning-at-home/05-the-block-and-the-stack/ https://anandpatel.com/series/fine-tuning-at-home/05-the-block-and-the-stack/ Tue, 01 Sep 2026 12:04:00 GMT What surrounds attention: the other half of a block, and two pieces of plumbing that look like footnotes and are the reason a model this deep can be trained at all. Then the cache that actually fills your graphics card, and why a model that loaded fine dies an hour later. https://anandpatel.com/series/fine-tuning-at-home/06-from-logits-to-words/ https://anandpatel.com/series/fine-tuning-at-home/06-from-logits-to-words/ Tue, 01 Sep 2026 12:05:00 GMT The top of the stack, where a vector becomes an actual word. Softmax, the sampling settings that live outside the model and change its character more than people expect, and how generation knows to stop, which is also one of the commonest ways a fine-tune breaks. https://anandpatel.com/series/fine-tuning-at-home/07-what-fine-tuning-changes/ https://anandpatel.com/series/fine-tuning-at-home/07-what-fine-tuning-changes/ Tue, 01 Sep 2026 12:06:00 GMT It is real training and it moves the weights permanently. What it moves reliably is form. What it moves badly is fact, and getting those backwards produces a system that is fluent, confident, and wrong. Also the two free things to try first, one of which may be enough. https://anandpatel.com/series/fine-tuning-at-home/08-building-the-training-set/ https://anandpatel.com/series/fine-tuning-at-home/08-building-the-training-set/ Tue, 01 Sep 2026 12:07:00 GMT The part that decides the outcome, and the one most likely to be skipped because none of it looks like machine learning. Turning fourteen poems into a training file, why six are held back, and the hardest question here: what to write on the prompt side when your source material is answers with the questions missing. https://anandpatel.com/series/fine-tuning-at-home/09-reading-the-scripts/ https://anandpatel.com/series/fine-tuning-at-home/09-reading-the-scripts/ Tue, 01 Sep 2026 12:08:00 GMT The three Python files, read properly: what each block does, which lines decide whether it fits on your card, and which four settings are worth touching. Everything in them is short enough to read, which is the point of not using a framework. https://anandpatel.com/series/fine-tuning-at-home/10-running-the-training/ https://anandpatel.com/series/fine-tuning-at-home/10-running-the-training/ Tue, 01 Sep 2026 12:09:00 GMT Every command, from a bare virtual machine to a saved adapter. On a single GPU. What the loss curve is and is not telling you, and the four things that went wrong when I did it. A second track runs on about 6 GB if you do not have a big card. https://anandpatel.com/series/fine-tuning-at-home/11-judging-the-result/ https://anandpatel.com/series/fine-tuning-at-home/11-judging-the-result/ Tue, 01 Sep 2026 12:10:00 GMT Merging the adapter, serving it, and testing it against six poems it was never shown, compared against both the untouched model and a carefully prompted one. That last comparison is the one most write-ups quietly skip because it can make the whole exercise look unnecessary.