---
title: Nanbeige4.1-3B-GGUF
canonical_url: "https://www.modelscope.cn/models/csncode/Nanbeige4.1-3B-GGUF"
md_url: "https://www.modelscope.cn/models/csncode/Nanbeige4.1-3B-GGUF.md"
repository: csncode/Nanbeige4.1-3B-GGUF
chinese_name: nanbeige/Nanbeige4.1-3B-GGUF
last_updated: 2026-03-08
license: apache-2.0
base_model:
  - nanbeige/Nanbeige4.1-3B
base_model_relation: quantized
library_name:
  - gguf
  - pytorch
  - transformers
frameworks:
  - PyTorch
  - transformers
language:
  - en
  - zh
downloads: 341
stars: 4
tags:
  - llm
  - nanbeige
  - GGUF
---

# Nanbeige4.1-3B-GGUF

> Nanbeige4.1-3B-GGUF - csncode 在 ModelScope 开源的模型。量化自Nanbeige4.1-3B

csncode/Nanbeige4.1-3B-GGUF 是 ModelScope 魔搭社区上的机器学习模型，采用 apache-2.0 许可，基于 nanbeige/Nanbeige4.1-3B 构建。

- **Repository**: csncode/Nanbeige4.1-3B-GGUF
- **License**: apache-2.0
- **Base model**: nanbeige/Nanbeige4.1-3B
- **Tags**: llm, nanbeige, GGUF
- **Downloads**: 341
- **Stars**: 4
- **Last updated**: 2026-03-08

Source: https://www.modelscope.cn/models/csncode/Nanbeige4.1-3B-GGUF

---

这是Nanbeige的GGUF量化版本。

原模型：[nanbeige/Nanbeige4.1-3B](https://modelscope.cn/models/nanbeige/Nanbeige4.1-3B)

量化工具：llama-quantize

## 使用方法

### 下载模型

根据您的设备情况下载合适的模型。

### 使用ollama运行模型

您需要下载q8_0版本及Modelfile文件，其他版本需修改Modelfile的FROM行。

Modelfile:
```
FROM Nanbeige4.1-3B-q8_0.gguf
TEMPLATE "{{- if .System }}<|im_start|>system
{{ .System }}<|im_end|>
{{ end -}}<|im_start|>user
{{ .Prompt }}<|im_end|>
<|im_start|>assistant
{{ .Response }}<|im_end|>
"
SYSTEM "You are Nange Pavilion, a professional large language model independently trained by BOSS Zhipin.

When responding to user messages you should strictly follow these guidelines:
- Do not overthink your response.
- Keep your thinking process simple and short.
- Think only as much as necessary to correctly answer the user.
- Avoid overthinking on unnecessary edge cases, excessive disclaimers, or speculative analysis.
- Do not overthink the instruction about not overthinking.
- Reconsider your train of thought only if there is a strong and clear reason to do so.
- Make reasonable assumptions if needed and proceed.
- Keep responses concise but complete.
- Stop once the core question is answered.
- Do not repeat the question.

Your objective is clarity, efficiency, and decisive execution — not exhaustive reasoning."
PARAMETER repeat_penalty 1
PARAMETER stop <|im_end|>
PARAMETER stop <|im_start|>
PARAMETER temperature 0.3
PARAMETER num_ctx 8192
```

当然，您也可以通过[](https://ollama.com/nuominecraft/nanbeige4.1-3B)来pull模型。