Smaug-Decisions

abacusai 's Collections

updated about 18 hours ago

Jev-type calibrated decision models for audio, images and text. Ahead of Gemini 4 Argon on audio decisions; runs offline on phones.

Running

1

On-Device Cascade

📱

1

Two decision models, one app, running on the phone

Note State-of-the-art audio decisions, offline on a phone
abacusai/Smaug-Decisions

Any-to-Any • 9B • Updated about 21 hours ago • 2

Note Smaug-Decisions: main model, 4B effective. Audio 60.1, text 56.6, images 57.9.
abacusai/Smaug-Audio-Decisions-0.7B

Audio Classification • 0.7B • Updated about 13 hours ago • 2 • 2

Note Fast audio tier, 744M. Audio 53.0, 389 ms on an iPhone 14 Pro Max.
abacusai/Smaug-Decisions-LiteRT

Any-to-Any • Updated about 13 hours ago • 2

Note Smaug-Decisions-LiteRT:Smaug-Decisions as one 3.45 GB LiteRT-LM file, runs on the phone GPU (iPhone and Android). Audio 59.4, text 53.9.