small language model · a 10,000‑parameter GPT that tokenizes, trains and generates entirely in this page
params
arch
vocab
ctx
step0
loss
steps/s0

training text

corpus
gradient descent on next-token prediction
loss
perplexity
|grad|
tokens0
sample generated while training

tokenization

byte-pair encoding
text → integers
The model never sees letters. A BPE tokenizer is learned from the corpus: start from single characters, then repeatedly merge the most frequent adjacent pair.
1 · vocabulary
2 · learned merges
3 · prompt, encoded
4 · token frequency in corpus

transformer

every intermediate tensor of one live forward pass
value scale −max  0  +max

weights & biases

one pixel per parameter · hover to identify
tensors
selected tensor
architecture

input

prompt
type to run a forward pass
temperature0.85
top-k10
max tokens140
Every keystroke runs a fresh forward pass, so the transformer panel always shows what the model is doing with the text above.

temperature reshapes the distribution before sampling — 0 always takes the most likely token, higher values flatten the odds. top-k throws away everything outside the k best candidates.

output

0 tokens
sampled one token at a time
next-token distribution