A small end-to-end LLM pipeline

LLM systems · Nanochat Study

A study implementation spanning tokenizer training, pretraining, supervised fine-tuning and reinforcement learning.

Question

Which parts of an LLM training pipeline become clearer when the whole path is kept small enough to inspect?

What the experiment produced

The repository connects BPE, pretraining, SFT and RL as one learning-oriented pipeline rather than isolated notebooks.

Inspect the work

This is a documented slice of Nanochat Study. The source repository is the authoritative implementation.