The model emits one byte.
U+000A, one byte, on demand. Prompts in, line feeds out, nothing else.
specimen 0autf-8 / 0a
U+000A / line feedadvance / 1 line
The model card, unabridged.
lf-1 has no weights, no tokenizer and no context window worth the name. What it has is an advance: one line, every time, in the same direction.
model card / lf-1rev. 0a
- parameters
- 1
- architecture
- a cursor
- input
- any
- output
- U+000A 100% of tokens
- context
- the current line
- temperature
- 0.0
- training data
- the return key
Five models. One output.
Every model in the catalog ends the line. Names are struck through for addresses the reputation service has marked dirty; the list stays visible so you know what you are missing.
5 models / 0 readablereason / E_IP_REPUTATION
Bring a prompt. Leave with a line.
Keys are issued to addresses with a clean reputation. The check runs before registration, before sign-in and before the catalog loads, so there is no form to fill in and nothing to confirm.
no email required / no card required / no access
POST /v1/lines200 OK
$ curl -s https://ai.0x0a.ru/v1/lines \ -H "authorization: Bearer $KEY" \ -d '{"model":"lf-1","input":"hello"}' | xxd 00000000: 0a .
the response, in full. shown for a clean address.
fig. 2: 4,096 blank lines, abridged.
end of file