Instructions to use Luna4682/Sentence2json with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use Luna4682/Sentence2json with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Luna4682/Sentence2json:Q8_0
# Run inference directly in the terminal:
llama cli -hf Luna4682/Sentence2json:Q8_0

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Luna4682/Sentence2json:Q8_0
# Run inference directly in the terminal:
llama cli -hf Luna4682/Sentence2json:Q8_0

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf Luna4682/Sentence2json:Q8_0
# Run inference directly in the terminal:
./llama-cli -hf Luna4682/Sentence2json:Q8_0

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf Luna4682/Sentence2json:Q8_0
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Luna4682/Sentence2json:Q8_0

Use Docker

docker model run hf.co/Luna4682/Sentence2json:Q8_0

LM Studio
Jan

vLLM

How to use Luna4682/Sentence2json with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Luna4682/Sentence2json"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Luna4682/Sentence2json",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Luna4682/Sentence2json:Q8_0

Ollama
How to use Luna4682/Sentence2json with Ollama:
```
ollama run hf.co/Luna4682/Sentence2json:Q8_0
```
Unsloth Desktop

How to use Luna4682/Sentence2json with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Luna4682/Sentence2json:Q8_0

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "Luna4682/Sentence2json:Q8_0"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use Luna4682/Sentence2json with Docker Model Runner:
```
docker model run hf.co/Luna4682/Sentence2json:Q8_0
```

Lemonade

How to use Luna4682/Sentence2json with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Luna4682/Sentence2json:Q8_0

Run and chat with the model

lemonade run user.Sentence2json-Q8_0

List all available models

lemonade list

Hermes Agent

How to use Luna4682/Sentence2json with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Luna4682/Sentence2json:Q8_0

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default Luna4682/Sentence2json:Q8_0

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use Luna4682/Sentence2json with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Luna4682/Sentence2json:Q8_0

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "Luna4682/Sentence2json:Q8_0" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Sentence2json — 中文单句 → 结构化事实（GGUF, Q8_0）

1.说明

这是一个把用户的一句话抽取成结构化事实的中文小模型；输出严格 JSON 行，只做"抽事实"这一件事，不做对话一般用于辅助Character.AI 类、AI 伴侣、视觉小说 NPC、桌宠、语音助手的人格记忆本模型完全在本地运行，不联网。您的私密对话、信息不会上传到云服务器

基座：Qwen2.5-1.5B-Instruct 量化：Q8_0

文件：fact-r8-q8_0.gguf（模型本体） system_prompt.txt（系统提示词，不可更换，否则指标会崩） SHA256SUMS.txt（校验码） README.txt（本文件）

2.用法

推荐使用llama.cpp（也支持Ollama，可自行选择）

示例：先确认llama-server可用起服务： llama-server -m fact-r8-q8_0.gguf --host 127.0.0.1 --port 8091 -c 2048 -np 1 -ngl 99 --alias r8

调用（OpenAI 兼容）：注意：一次请求发一句话，不要成批输入 curl http://127.0.0.1:8091/v1/chat/completions -H "Content-Type: application/json" -d '{ "model": "r8", "temperature": 0, "max_tokens": 256,"repeat_penalty": 1.0, "messages": [ {"role": "system", "content": "<system_prompt.txt 全文>"}, {"role": "user", "content": "<你要输入的原文>"} ] }'

注意：调用时temperature一律建议手动设为0，否则模型会明显退化 max_tokens设256够用，经测试99%的输出在60token内 repeat_penalty建议设为1.0，调高可能会出现坏json（若你的客户端不认这个字段，可在启动参数里加 --repeat-penalty 1.0，两种写法等效。） repeat_penalty 为 llama.cpp 扩展参数，换到其它后端时删除该项即可

3.输出格式

一行一个 JSON 对象。共三种形态：

// 形态 1：不自足（指代 / 省略 / 确认词 / 废话） {"has_info": false, "self_contained": false}

// 形态 2/3：自足但没有关于"我"的可记信息（第三人称、反问、评价、玩笑…） {"has_info": false, "self_contained": true, "certainty": "explicit"}

// 正例：全部字段 {"has_info": true, "self_contained": true, "subject_raw": "我", "predicate": "对", "object_raw": "花生过敏", "polarity": 1, "kind": "attribute", "certainty": "explicit", "time_expression": null}

字段：has_info, self_contained, subject_raw, predicate, object_raw, polarity, kind, certainty, time_expression

4.评测

测试时采用真实聊天记录289条（大量指代不明，省略）+手写句子100条（大多是完整句子） r8：真实记录289，含27条正事实：json合法率100%，负事实判正率0.8%（2条），正事实丢失率25.9%（7条），字段准确率95.7% 手写句子100，含56条正事实：json合法率100%，负事实判正率0%（0条），正事实丢失率5.4%（3条），字段准确率98.9% 本地Vulkan，单句延迟≈0.65s，单句成本0

对比DeepSeekv4.1flash（相同提示词，额外提供1419字符规则文本）：真实记录289，含27条正事实：json合法率100%，负事实判正率14.1%（37条），正事实丢失率7.4%（2条），字段准确率86.9% 手写句子100，含56条正事实：json合法率100%，负事实判正率31.8%（14条），正事实丢失率1.8%（1条），字段准确率85.5% 单句延迟≈1.0s（含网络），单句成本约 800 tokens

对比Qwen3.8Max（相同提示词，额外提供1419字符规则文本）：真实记录289，含27条正事实：json合法率100%，负事实判正率7.3%（19条），正事实丢失率22.2%（6条），字段准确率88.4% 手写句子100，含56条正事实：我没跑，太贵了（目移）跑完289条就花了十块多单句延迟≈10.8s（含网络），单句成本约 1400 tokens

5.已知限制

中文为主，支持简单英文（但未系统实测）设计为“只记录与说话人直接相关的、有记录价值的事实，拿不准就判负”，尽量保持数据入库时“宁可漏不可错” 为了方便抽取，设计为只抽取句子中出现的原话（除部分句式可以补“是”“有”来保证谓语行非空缺），但经过实测，模型仍有极小概率自己造字，不可完全信赖。使用时需要有后端逻辑兜底本模型每次只抽取第一个关于说话人本身的事实分句，多事实句会舍弃超出第一个的部分，如有需要，请将句子切分后再喂给模型在第一人称自述上训练：第三人称、反问、元话语、纯情绪会被判负，这是设计而非缺陷。

6.许可

权重与基座同为 Apache-2.0（Qwen2.5-1.5B-Instruct LICENSE：https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct/blob/main/LICENSE）训练数据为自建中文语料，不含基座模型输出以外的第三方数据。

7.字段含义

"has_info": {"type": "boolean", "desc": "这句话里有没有关于“我”的可记事实"}, "self_contained": {"type": "boolean", "desc": "单拿出来看不看得懂（看不懂=不自足）"}, "subject_raw": {"type": "string", "desc": "只写第一人称（我/我们/咱们…），不还原、不补全"}, "predicate": {"type": "string", "desc": "从原句取；必要时补“是”“有”；体标记（了/过/着）紧跟谓语就归谓语"}, "object_raw": {"type": "string|null", "desc": "谓语后到第一个分句标点为止；没有就 null"}, "polarity": {"type": "integer", "enum": [1, -1],"desc": "肯定为1，否定为-1"}, "kind": {"type": "string", "enum": ["preference", "attribute", "event", "commitment", "relationship"]}, "certainty": {"type": "string", "enum": ["explicit", "speculation", "joke"]}, "time_expression": {"type": "string|null", "desc": "时间/频率的原话，不归一化"}

kind字段解释： preference：偏好，喜好 attribute：状态，习惯，属性，特质 event：做过的事情 commitment：将要做的事情 relationship：拥有，持有关系注意：很多句子的kind比较特殊，无法准确归类，因此该字段只能用作参考

8.简单单句示例

in : 我对花生过敏 out: {"has_info": true, "self_contained": true, "subject_raw": "我", "predicate": "对","object_raw": "花生过敏", "polarity": 1, "kind": "attribute","certainty": "explicit", "time_expression": null}

in : 我有个弟弟 out: {"has_info": true, "self_contained": true, "subject_raw": "我", "predicate": "有","object_raw": "个弟弟", "polarity": 1, "kind": "relationship","certainty": "explicit", "time_expression": null}

in：我不太喜欢公司的氛围 out：{"has_info": true, "self_contained": true, "subject_raw": "我", "predicate": "喜欢", "object_raw": "公司的氛围", "polarity": -1, "kind": "preference", "certainty": "explicit", "time_expression": null}

in : 你连这都知道，真牛逼啊 out: {"has_info": false, "self_contained": false}

in : 不，我的意思是，你知道这意味着什么吗 out: {"has_info": false, "self_contained": false}

in : 我昨天还想今天去医院来着，不过今天突然说是有课 out: {"has_info": false, "self_contained": true, "certainty": "explicit"}

Downloads last month: 9

GGUF

Model size

2B params

Architecture

qwen2

Hardware compatibility

8-bit

Model tree for Luna4682/Sentence2json

Base model

Qwen/Qwen2.5-1.5B

Finetuned

Qwen/Qwen2.5-1.5B-Instruct

Quantized

(332)

this model