systemone-lite-0.5b
A 0.5B decision model on Qwen2.5, the only entry here built on a previous-generation base. Real downloads, zero likes.
Community card
What it is
A 0.5B decoder on Qwen2.5, aimed at the same target as
rlcd-modernbert-151m: a decision model small enough to run
next to the thing making the decision. It is the only model in the directory built on a
previous-generation base, where the rest of the decoder-based entries went to Qwen3.5.
Plenty of downloads, no likes
An odd shape, and worth not over-reading. It can mean people pulled it, tried it and moved on. It can equally mean an automated mirror or a training pipeline pulling weights on a schedule, which produces downloads and never produces likes. Hugging Face counters do not distinguish between a human and a cron job, which is a decent argument for why download counts should not be used as quality signals at all.
What it is useful for finding out
The interesting comparison is not this model against Jev, it is this model against
rlcd-modernbert-151m. Both are sub-1B, both are trying to be the cheap tier, and they take
opposite architectural routes: a small decoder that generates its answer versus a small encoder
that classifies. Whichever wins at this size tells you which route the cheap tier of this market
is going to take.
A public number is a shortlist, not a decision
The only benchmark that settles which model you ship is one built from your own historical decisions. We freeze that test set, run the candidates against it, and hand back the threshold your error budget supports.