Machine Learning

Ever wondered how a computer learns to generate text that actually looks like Malayalam? Not just random characters, but something with actual structure? I’m not talking about Large Language Models here. I’m talking about Small Language Models that are efficient and explainable. something you can build and run on your own laptop. In my previous “The Broken Token” article, I presented a Malayalam unigram tokenizer and analysed its strengths and weaknesses. I did fertility rate evaluation and then analysed the tokenization in the context of Malayalam language characteristics. A common evaluation method for tokenizers is using them in downstream tasks—so I decided to build a text generator. That’s where things got interesting. ...

Tokenization is the first cut a language model makes on text. For Malayalam, that cut has almost always been in the wrong place — and the consequences reach much further than efficiency. Most large language models use tokenizers trained primarily on English and related Latin-script languages. When those tokenizers encounter Malayalam, the results range from inefficient to broken. Andrej Karpathy captured this as follows: This post is in two connected parts. The first is a technical exercise: I trained two subword tokenizers for Malayalam — one BPE, one Unigram — on the SMC Malayalam corpus and published them on HuggingFace. I explain the algorithms, the design choices that make them suitable for Malayalam, and measure their fertility against a cross-section of major LLMs including GPT-4, Gemma, Sarvam, and LLaMA. ...