0xSojalSec
A pseudonymous GitHub curator who maintains catalogues of local, free, open-weight models, one per modality. Paged now that the same maintainer has surfaced in a second spoke — the hub’s rule for deferred thin creators is to page them when one recurs.
The evidence, which is all of it
Two repositories, both README-only awesome-lists with the same construction (comparison table, per-entry detail block, GitHub and Hugging Face badges, licence column):
- free-voice-clone — 53 speech and audio models across TTS, music, anything-to-audio, restoration and ASR. Held here as free-voice-clone-list. 450★, last pushed 2026-04-09.
- LLMs-local — the on-device text-LLM stack: platforms, engines, UIs, tutorials. Held in llm-inference-wiki as llms-local-list, where it was ingested via a Dan Kornas tweet.
Nothing else is known here: no name, no affiliation, no stated expertise, no published methodology for either list, and no licence on either repo. Both entered this hub as T4 sources, and the tier is about the absence of process, not about the accuracy of any individual row.
Why the recurrence is worth a node
The two lists are the same editorial act performed on two modalities: catalogue what you can run yourself, filter on licence and footprint rather than on benchmark rank. Both organise by capability and licence and neither publishes a measurement. That is a genre — the local-first model census — and this curator is now the hub’s clearest instance of it, spanning text-to-speech and text inference alike.
The genre’s characteristic weakness travels with it: coverage without evaluation, and, in the voice case, coverage without any mention of consent (free-voice-clone-list).
Related
free-voice-clone-list · llms-local-list · open-weight-tts · text-to-speech