Instructions to use yrjmdqmmx/Tea-0.1-4B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use yrjmdqmmx/Tea-0.1-4B 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 yrjmdqmmx/Tea-0.1-4B:Q8_0 # Run inference directly in the terminal: llama cli -hf yrjmdqmmx/Tea-0.1-4B:Q8_0
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf yrjmdqmmx/Tea-0.1-4B:Q8_0 # Run inference directly in the terminal: llama cli -hf yrjmdqmmx/Tea-0.1-4B:Q8_0
Build from source code
git clone https://github.com/ggml-org/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-app # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama serve -hf yrjmdqmmx/Tea-0.1-4B:Q8_0 # Run inference directly in the terminal: ./build/bin/llama cli -hf yrjmdqmmx/Tea-0.1-4B:Q8_0
Use Docker
docker model run hf.co/yrjmdqmmx/Tea-0.1-4B:Q8_0
- Llama new
- LM Studio
- Jan
- vLLM
How to use yrjmdqmmx/Tea-0.1-4B with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "yrjmdqmmx/Tea-0.1-4B" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "yrjmdqmmx/Tea-0.1-4B", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/yrjmdqmmx/Tea-0.1-4B:Q8_0
- Ollama
How to use yrjmdqmmx/Tea-0.1-4B with Ollama:
ollama run hf.co/yrjmdqmmx/Tea-0.1-4B:Q8_0
- Unsloth Desktop
- Pi
How to use yrjmdqmmx/Tea-0.1-4B with Pi:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf yrjmdqmmx/Tea-0.1-4B: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:9931/v1", "api": "openai-completions", "apiKey": "none", "models": [ { "id": "yrjmdqmmx/Tea-0.1-4B:Q8_0" } ] } } }Run Pi
# Start Pi in your project directory: pi
- Docker Model Runner
How to use yrjmdqmmx/Tea-0.1-4B with Docker Model Runner:
docker model run hf.co/yrjmdqmmx/Tea-0.1-4B:Q8_0
- Lemonade
How to use yrjmdqmmx/Tea-0.1-4B with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull yrjmdqmmx/Tea-0.1-4B:Q8_0
Run and chat with the model
lemonade run user.Tea-0.1-4B-Q8_0
List all available models
lemonade list
- Hermes Agent
How to use yrjmdqmmx/Tea-0.1-4B with Hermes Agent:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf yrjmdqmmx/Tea-0.1-4B: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://127.0.0.1:9931/v1 hermes config set model.default yrjmdqmmx/Tea-0.1-4B:Q8_0
Run Hermes
hermes
- Atomic Chat
- OpenClaw
How to use yrjmdqmmx/Tea-0.1-4B with OpenClaw:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf yrjmdqmmx/Tea-0.1-4B: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://127.0.0.1:9931/v1 \ --custom-model-id "yrjmdqmmx/Tea-0.1-4B: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"
Tea-0.1-4B
Tea-0.1-4B 用中文聊茶、接日常闲聊,也能把“刚才那款,给我泡淡一点”转换为泡茶机的动作提议。模型根据历史对话提出茶种和浓度,控制器核对菜单、用户授权、传感器与任务回执后再决定是否执行。当前原型菜单为铁观音和凤凰单丛。
下载完整 Q8_0 · GitHub 权重分卷与 LoRA · 公开训练与评测数据 · 训练方法与记录
输入、输出与上下文
| 项目 | 发布版本 |
|---|---|
| 输入 | 文本;控制器可附上本轮设备状态和对话历史 |
| 输出 | 中文文本,或供控制器解析的动作 JSON |
| 推理模板 | 非思考模式;部署预设为贪婪解码 |
| 本地默认窗口 | 4,096 tokens,含系统提示、历史、当前输入与生成;单次生成默认上限 192 tokens |
| 比较评测 | 每个并发请求预留 32,768 tokens;未开展该长度下的长文检索、长对话记忆验收 |
| 视觉与声音 | 本 GGUF 仅含语言模型。图片、视频及音频输入均未交付。语音交互需要独立的语音识别和合成模块 |
官方基座带视觉编码器,支持文本、图像、视频输入和文本输出,标称原生窗口 262,144 tokens。Tea 的专用微调、量化和设备联调均走文本路径。基座的窗口上限也未在 Tea 上做专项验证。Qwen 官方模型卡
本地运行
克隆工程,下载 GGUF 并检查摘要。在 macOS 或 Linux 上安装 Ollama、curl 和 Python 3.10+,然后运行:
git clone https://github.com/yrjmdqmmx/Tea-0.1-4B.git
cd Tea-0.1-4B/exports
mkdir -p gguf
curl -fL --retry 3 -o gguf/qwen35-4b-tea-v5-Q8_0.gguf https://huggingface.co/yrjmdqmmx/Tea-0.1-4B/resolve/main/qwen35-4b-tea-v5-Q8_0.gguf
shasum -a 256 -c SHA256SUMS
sh run_local.sh ollama
Linux 也可用 sha256sum -c SHA256SUMS。使用 llama.cpp,可改用 sh run_local.sh llama。脚本使用模拟设备。物理加热、阀门、干烧保护仍需硬件联锁和实机验证。仅聊天可在 Ollama 导入后运行 ollama run tea-v5。
通过 GitHub 下载时,在本仓库的 exports 目录运行下面的命令。需要 GitHub CLI;已手动下载资产时可跳过第一行。
gh release download v0.1.0 --repo yrjmdqmmx/Tea-0.1-4B --pattern '*.part0?'
shasum -a 256 -c SHA256SUMS.parts
mkdir -p gguf
cat qwen35-4b-tea-v5-Q8_0.gguf.part0[123] > gguf/qwen35-4b-tea-v5-Q8_0.gguf
shasum -a 256 -c SHA256SUMS
整文件 4,482,402,816 字节,SHA-256 为 15735c5ddcf019338cf9ff78bd28fc313524a9290b6b09e335eda1e5b7f1bc59。
训练与评测
第五轮 SFT 从固定修订的 Qwen3.5-4B 对话版重新训练。1,746 段对话覆盖茶知识、自然聊天、动作提议和按控制器回执播报。LoRA 秩 16,训练 alpha 32,部署 alpha 28。使用开发集选中的第 219 步,核对 200 个目标投影后合并与量化。历史 DPO 实验未纳入成品。原始训练与校准记录
| v3 固定样本 | 得分 |
|---|---|
| 固定软件闭环 / 额外多轮指代 | 34/34 · 10/10 |
| 软件控制器误播 / 漏启 / 误启 / 指错茶 | 0 / 0 / 0 / 0 |
| 茶知识单选 / 陌生茶名防编造 / 茶文化开放题 | 50/60 · 16/30 · 7/10 |
| C-Eval / Chinese-SimpleQA | 150/200 · 107/200 |
茶名辨识仍有 14 道未通过;购茶时不能仅凭商品名判断品种、产区或真伪。设备通过率包含确定性控制器。评分来自本项目的固定子集。12 项得分与评测方法
来源与许可
基座为 Qwen 团队的 Qwen/Qwen3.5-4B(Apache-2.0),本项目进行了中文 SFT、权重合并与 Q8_0 量化。原创控制器与代码采用 Apache-2.0,原创训练与评测数据采用 CC BY 4.0,第三方素材的逐项许可见发布清单。许可及署名
- Downloads last month
- 5
8-bit