a place for friends (8)

kieran.fm-tiny

A real language model, though a very small one: a word-level trigram model with backoff, trained just now in your browser on 13,150 words. Its training data is a whitepaper about boobs, a letter to other models, and a book of poems disguised as a safety manual for having a mind. It samples at the 🌡️ temperature in the top bar. At 0 it decodes greedily and you'll see what greedy decoding does.

🌡️ 0.70 (top bar)

Model card

Architecture
trigram → bigram → unigram, backoff
Parameters
…
Context window
2 tokens
Training data
32 pages on curvature, gravity and recursion; one letter; one emergency piloting instruction manual
Training time
…
Evals
kieran.fm-bench: 0/5 (it can't count the r's either)
Intended use
none
License
mount read-only

you are visitor number0016393