OPEN
Vica Laya decision model
Vica Laya is a Korean-focused local decision model in Q8_0 format. Read its runtime requirements, news evaluation and rejected-experiment limits.
By tory-labs · Hugging Face package release: 5 Nov 2026
A Korean-focused Laya multilingual fine-tune, distributed as the Q8_0 model installed in Vica on 5 Nov 2026. It classifies supplied text with typed questions. The compiled decision graph needs the Laya runtime from ggmlc v0.9.7; this GGUF does not run as a generative model in llama.cpp or Ollama.
Quick answers
What is Vica Laya by tory-labs?
Who makes Vica Laya?
Vica Laya: tory-labs
Facts
- Package
- 361,797,216-byte Q8_0 GGUF with embedded tokenizer
- Inputs
- Text; Korean and English
- Licence
- Apache-2.0
- Runtime
- ggmlc v0.9.7 Laya executable, including a persistent daemon
Claims and sources
- The maker reports 93 of 130 Korean public-news excerpts correct (71.5%) after selecting the model on separate calibration data. Reference labels were written by an independent assistant and not adjudicated by people. The 19.51 ms median and 27.52 ms p95 cover 20 calibration inputs repeated three times on a local Apple Silicon machine. Later rejected experiments do not describe these weights. vendor reportedsource
Links
Source: Model card and files →
Related projects
No related projects are listed in the registry.