Onnx-Library — try the whole family
Every model is an ONNX graph streamed from Quazim0t0/Onnx-Library and run on CPU. First use of each model downloads it, then it's cached.
Model
8 128
0 1.5
Examples
Byrne-ASR-English — upload/record English speech (greedy CTC; no LM rescoring).
Byrne-Speech — FastSpeech2 → HiFi-GAN, 24 kHz.
Examples
Vision encoders — image → 512-d embedding + 784 patch tokens (features, no text).
Encoder
Vision-language generation — upload an image and generate text. Byrne-VLM captions natural images; Byrne-Docling emits DocTags for document images. Runs the full vision→connector→LM pipeline on CPU, so generation is slow (~seconds/token).
Model
8 64
0 1.2