---
license: apache-2.0
library_name: llama.cpp
pipeline_tag: text-generation
language: [zh, en]
base_model:
  - Qwen/Qwen3.8-27B
  - prism-ml/Ternary-Bonsai-27B-gguf
tags: [gguf, reasoning, bounded-thinking, ternary, q2_0, llama-cpp, lm-studio, experimental]
inference: false
---

# Vireqo-27T-Plus-260818

`Vireqo-27T-Plus-260818` is the ternary bounded-Thinking product built on the unchanged `Vireqo-27B-Plus-260816` payload. It combines the Plus language operating point with the independently validated `Think512-Concise` LM Studio preset.

中文摘要：这是独立的 T-Plus 产品线，不是把 Thinking 直接塞回原 Plus 模型。原 Plus 权重、SHA 和产品定位保持不变；本目录通过软链接复用主权重。

## Bundle

| Item | Value |
|---|---|
| Main file | `Vireqo-27T-Plus-260818.gguf` |
| Physical source | unchanged `Vireqo-27B-Plus-260816.gguf` |
| Size | 7,585,332,288 bytes / 7.0644 GiB |
| SHA-256 | `a32a8ec286a11c6534bf29d1ee20bd4c02064032b51ae8310bb1216e2de17e03` |
| Preset | `thinking-preset.json` |
| New physical main-weight copy | no |

The internal `general.name` remains `Vireqo-27B-Plus-260816`; the T-Plus identity is supplied by the release bundle and LM Studio key.

## Accepted preset

Think On; Reasoning Budget `512`; maximum response `768`; temperature `0`; top-p `1`; repeat penalty `1.08`; context `2048`; parallel `1`.

Budget Message:

```text
思考预算已到。复核已有结论后，只输出最终答案；禁止重复或展示思考过程。
```

## Validation

Capital, multiplication, and chicken/rabbit final answers were all correct with separated nonempty reasoning and final content. See [`thinking-validation.md`](thinking-validation.md). Plus tends to use more of the 512-token reasoning budget and is slower than standard T.

## Limitations

- Unrestricted Thinking remains unsupported.
- This concise preset is not a general reasoning benchmark claim.
- Complex tasks may require a future larger-budget line.
- The underlying ternary model retains all Plus limitations.

See [`LM-STUDIO-使用指南.md`](LM-STUDIO-使用指南.md), [`TECHNICAL_README.md`](TECHNICAL_README.md), and [`bundle-provenance.json`](bundle-provenance.json).