Alexhf825/AutoMR-pangu

AutoMR for Pangu

This project equips the Pangu model with the AutoMR reasoning framework, optimized for Huawei Ascend hardware.

🌳 Project Structure

.
├── AMC.SH
├── automr
│   ├── config.py
│   ├── dag.py
│   ├── data_loader.py
│   ├── evaluator.py
│   ├── __init__.py
│   ├── model.py
│   ├── strategies.py
│   ├── trainer.py
│   └── utils.py
├── checkpoints
│   └── MATH
├── embedder_server.sh
├── generator_server.sh
├── main.py
├── math_train.sh
└── processed_data
    ├── AMC
    └── MATH

🔧 1. Installation

a. Clone the Project Repository

hf download Alexhf825/AutoMR-pangu --local-dir AutoMR-pangu
cd AutoMR-pangu

b. Install Dependencies

c. Download Datasets

This command will download the datasets and place them in the ./processed_data directory, matching the project structure.

hf download Alexhf825/dataset-AutoMR-pangu --repo-type=dataset --local-dir=./

🚀 2. Start the Servers

This project requires two services running in an OpenAI-API style. Please run the following commands in two separate terminal sessions.

Start the Embedder Server:

bash embedder_server.sh

Start the Generator Server:

bash generator_server.sh

📈 3. Run Evaluation

A pre-trained checkpoint (MATH) is provided. You can directly evaluate the model on the AMC dataset using the following command:

bash AMC.sh

🏋️ 4. Run Training

You can also train the model from scratch on the MATH dataset by running:

bash math_train.sh