# Chinh Nguyen: quyet

> Quyet 1.0 is Chinh Nguyen's Apache-2.0 decision-model family: it reads a state and answers choice, score and noul questions with calibrated probabilities over the options given, generating no text. This entry is Large, a merged LoRA fine-tune of Gemma-4-31B-it.

- Page: https://systemonemodels.ai/chinh-nguyen/quyet
- API: https://api.systemonemodels.ai/v1/models/chinh-nguyen/quyet
- Download: `pip install systemonemodels && systemone pull chinh-nguyen/quyet`

## Facts

| | |
|---|---|
| Maker | Chinh Nguyen (https://systemonemodels.ai/chinh-nguyen) |
| Decides | choice, score, noul |
| Architecture | quyet |
| Base model | google/gemma-4-31b-it |
| Parameters | 31B |
| Context | 8K tokens |
| Licence | apache-2.0 |
| Availability | Open weights |
| Released | 2026-10-03 |
| Latest version | 1.0.0 |

## Model card

<!-- generated by scripts/seed_catalog.py; edit content/models/catalog.yaml -->

# Quyet-1.0-Large

Quyet 1.0 is Chinh Nguyen's Apache-2.0 decision-model family: it reads a state and answers choice, score and noul questions with calibrated probabilities over the options given, generating no text. This entry is Large, a merged LoRA fine-tune of Gemma-4-31B-it.

A personal release by Chinh Nguyen (GitHub ncchinh, Hugging Face chinhnc), published on 3 October 2026 with a pip runtime (`pip install quyet`, `quyet.load(...)`, plus a `quyet predict` command line). Large is Gemma-4-31B-it with a merged rank-16 LoRA; it answers by reading the next-token probabilities of the option letters (A, B, ...) after a fixed prompt that the package builds, and applies per-type temperatures from quyet_config.json. It loads as a plain Gemma 4 checkpoint in transformers or vLLM, but without the package's prompt and temperatures it is just the chat model. bf16 weights are 62.5 GB and need one 80 GB GPU or several with device_map="auto". The state is capped at 6,000 tokens inside an 8,000-token prompt, and a question takes at most 10 options. Tuned for English and Vietnamese. Siblings, all Apache-2.0: Medium (Qwen3.5-4B, 4.66B), and three CPU-friendly encoders that score every option in one 8,192-token pass, Small (SEA-LION-ModernBERT-300M, 328M), Small-EN (ModernBERT-base, 153M, English only) and Tiny (mmBERT-small cut to 16 layers, 183M, distilled from Medium); the encoders' option rendering follows Convai's Laya design, with no Laya weights. The maker publishes no benchmark results, no calibration measurements and no description of the training data, and says the models are not evaluated for safety, bias or adversarial robustness. On the third-party Decision Index 0.3 (multimodalart's Hugging Face Space, 6 October 2026) Large scores 58.71, seventh of 112 entries (60.79 on the public part). Redistribution must keep the NOTICE file, which credits "Quyet by Chinh Nguyen".

## What it decides

- **choice** — picks one option from a set
- **score** — places the input on an ordered scale
- **noul** — answers a yes/no question with one probability

## At a glance

| | |
|---|---|
| Parameters | 31.3B |
| Base model | `google/gemma-4-31B-it` |
| Maker | Chinh Nguyen |
| Released | 2026-10-03 |
| License | apache-2.0 |

## Get the weights

```bash
pip install systemonemodels
systemone pull chinh-nguyen/quyet
```

The files are served from the maker's Hugging Face repository, [`chinhnc/Quyet-1.0-Large`](https://huggingface.co/chinhnc/Quyet-1.0-Large), and verified against the checksums recorded here.

## Read more

- [Model card](https://huggingface.co/chinhnc/Quyet-1.0-Large)
- [Runtime and source (GitHub)](https://github.com/ncchinh/quyet)
- [pip package](https://pypi.org/project/quyet/)
- [Quyet-1.0-Medium](https://huggingface.co/chinhnc/Quyet-1.0-Medium)
- [Quyet-1.0-Small](https://huggingface.co/chinhnc/Quyet-1.0-Small)
- [Quyet-1.0-Small-EN](https://huggingface.co/chinhnc/Quyet-1.0-Small-EN)
- [Quyet-1.0-Tiny](https://huggingface.co/chinhnc/Quyet-1.0-Tiny)

---

*This page was opened by System One for Chinh Nguyen, who can claim the organisation and take it over at any time.*

---

From System One Models — https://systemonemodels.ai/ · every System One model: https://systemonemodels.ai/system-one-models