Instructions to use lovedheart/openPangu-2.0-Flash-GGUF-Preview with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use lovedheart/openPangu-2.0-Flash-GGUF-Preview 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 lovedheart/openPangu-2.0-Flash-GGUF-Preview:MXFP4_MOE
# Run inference directly in the terminal:
llama cli -hf lovedheart/openPangu-2.0-Flash-GGUF-Preview:MXFP4_MOE

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf lovedheart/openPangu-2.0-Flash-GGUF-Preview:MXFP4_MOE
# Run inference directly in the terminal:
llama cli -hf lovedheart/openPangu-2.0-Flash-GGUF-Preview:MXFP4_MOE

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 lovedheart/openPangu-2.0-Flash-GGUF-Preview:MXFP4_MOE
# Run inference directly in the terminal:
./llama-cli -hf lovedheart/openPangu-2.0-Flash-GGUF-Preview:MXFP4_MOE

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 lovedheart/openPangu-2.0-Flash-GGUF-Preview:MXFP4_MOE
# Run inference directly in the terminal:
./build/bin/llama-cli -hf lovedheart/openPangu-2.0-Flash-GGUF-Preview:MXFP4_MOE

Use Docker

docker model run hf.co/lovedheart/openPangu-2.0-Flash-GGUF-Preview:MXFP4_MOE

LM Studio
Jan
Ollama
How to use lovedheart/openPangu-2.0-Flash-GGUF-Preview with Ollama:
```
ollama run hf.co/lovedheart/openPangu-2.0-Flash-GGUF-Preview:MXFP4_MOE
```

Unsloth Studio

How to use lovedheart/openPangu-2.0-Flash-GGUF-Preview with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for lovedheart/openPangu-2.0-Flash-GGUF-Preview to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for lovedheart/openPangu-2.0-Flash-GGUF-Preview to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for lovedheart/openPangu-2.0-Flash-GGUF-Preview to start chatting

How to use lovedheart/openPangu-2.0-Flash-GGUF-Preview with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf lovedheart/openPangu-2.0-Flash-GGUF-Preview:MXFP4_MOE

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "lovedheart/openPangu-2.0-Flash-GGUF-Preview:MXFP4_MOE"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use lovedheart/openPangu-2.0-Flash-GGUF-Preview with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf lovedheart/openPangu-2.0-Flash-GGUF-Preview:MXFP4_MOE

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 lovedheart/openPangu-2.0-Flash-GGUF-Preview:MXFP4_MOE

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use lovedheart/openPangu-2.0-Flash-GGUF-Preview with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf lovedheart/openPangu-2.0-Flash-GGUF-Preview:MXFP4_MOE

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 "lovedheart/openPangu-2.0-Flash-GGUF-Preview:MXFP4_MOE" \
  --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"

Docker Model Runner
How to use lovedheart/openPangu-2.0-Flash-GGUF-Preview with Docker Model Runner:
```
docker model run hf.co/lovedheart/openPangu-2.0-Flash-GGUF-Preview:MXFP4_MOE
```

Lemonade

How to use lovedheart/openPangu-2.0-Flash-GGUF-Preview with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull lovedheart/openPangu-2.0-Flash-GGUF-Preview:MXFP4_MOE

Run and chat with the model

lemonade run user.openPangu-2.0-Flash-GGUF-Preview-MXFP4_MOE

List all available models

lemonade list

openPangu-2.0-Flash

中文 | English

1. 简介

openPangu-2.0-Flash 是基于昇腾 NPU 训练的大规模混合专家（MoE）语言模型，参数规模约 92B，每 token 激活参数规模约 6B，模型支持 512k 上下文长度，训练数据总量约34T tokens。后训练阶段完成快慢合一微调（SFT）、多专项强化学习（RL），并通过在线蒸馏（OPD）完成能力合一。

2. 模型架构

openPangu-2.0-Flash 在模型架构上实现了全面的升级：

Attention架构：沿用高效 MLA，并采用 DSA+SWA 独立分层混合架构，层配比为 1:2；SWA 层负责局部窗口建模，DSA 层负责稀疏全局聚合，在保持精度的同时显著降低长序列推理的计算、显存与访存开销。
拓扑架构：将传统残差连接升级为 4 支流 mHC 架构，提升表征多样性与泛化能力。
自投机模块：采用 3 头 MTP 架构，一次额外预测 3 个 token，显著提升模型的推理速度。
Muon优化器：训练中采用Muon优化器，获得更快的收敛速度。

3. 测评结果

测评集	测评指标	openPangu-2.0-Flash-Thinking	openPangu-2.0-Flash-Non-Thinking
通用能力
CL-Bench	Acc	20.4	15.5
IFEval	Prompt Strict	95.9	89.3
IFBench	Prompt Strict	79.6	54.4
AgentIF	(CSR+ISR)/2	44.9	43.9
SysBench	ISR	91.1	87.9
Multichallenge	Acc	68.4	51.9
推理能力
AIME 2026	Avg@16	93.3	86.5
- w/ Python	Avg@16	98.1	-
HMMT Feb 2025	Avg@16	91.5	67.1
- w/ Python	Avg@16	96.9	-
IMO-AnswerBench	Acc	76.5	62.3
- w/ Python	Acc	80.8	-
BBEH	Harmonic Mean	62.5	51.5
GPQA-Diamond	Avg@4	83.7	79.8
Agent能力
BrowseComp	Acc	57.0	-
SkillsBench	Avg@5	42.6	40.0
PinchBench	Avg@3	85.6	82.5
Claw-Eval	Pass^3	57.7	58.2
WildClawBench	Avg@3	41.5	35.0
MCP-Atlas	Acc	58.9	47.9
TAU2-Bench	Avg@3	88.0	74.0
代码能力
LiveCodeBench V6	Avg@3	85.1	50.9
DeepCodeBench	Avg@3	76.5	70.9
SWE-bench Verified	Avg@3	63.1	57.6
FeatBench	Avg@3	45.9	45.8

4. 部署和使用

使用omni-infer推理框架，部署方式参考[openPangu-2.0-Flash部署说明]
源码仓:[openPangu-2.0-Infer]

5. 模型许可证

除文件中对开源许可证另有约定外，openPangu-2.0-Flash 模型根据 OPENPANGU MODEL LICENSE AGREEMENT VERSION 2.0 授权，旨在允许使用并促进人工智能技术的进一步发展。有关详细信息，请参阅模型存储库根目录中的 LICENSE 文件。

6. 免责声明

由于 openPangu-2.0-Flash （“模型”）所依赖的技术固有的限制，以及人工智能生成的内容是由模型自动生成的，华为无法对以下事项做出任何保证：

该模型的输出通过AI算法自动生成，不能排除某些信息可能存在缺陷、不合理或引起不适的可能性，生成的内容不代表华为的态度或立场；
无法保证该模型100%准确、可靠、功能齐全、及时、安全、无错误、不间断、持续稳定或无任何故障；
该模型的输出内容不构成任何建议或决策，也不保证生成的内容的真实性、完整性、准确性、及时性、合法性、功能性或实用性。生成的内容不能替代医疗、法律等领域的专业人士回答您的问题。生成的内容仅供参考，不代表华为的任何态度、立场或观点。您需要根据实际情况做出独立判断，华为不承担任何责任。

7. 反馈

如果有任何意见和建议，请提交issue或联系openPangu@huawei.com。

Downloads last month: 91

GGUF

Model size

92B params

Architecture

openpangu-v2

Hardware compatibility

4-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for lovedheart/openPangu-2.0-Flash-GGUF-Preview

Base model

openpangu/openPangu-2.0-Flash

Quantized

(5)

this model