Skip to content

README_EN.md

Latest commit

 

History

History
129 lines (89 loc) · 3.56 KB

README_EN.md

File metadata and controls

129 lines (89 loc) · 3.56 KB

PAI Python SDK

English | 简体中文

The PAI Python SDK is provided by Alibaba Cloud's Platform for Artificial Intelligence (PAI). It offers a user-friendly High-Level API, enabling machine learning engineers to easily train and deploy models on PAI using Python, streamlining the machine learning workflow.

Installation 🔧

Install the PAI Python SDK using the following command, which supports Python versions >= 3.8 :

python -m pip install pai

📖 Documentation

Find detailed documentation, including API references and user guides, in the docs directory or visit PAI Python SDK Documentation.

🛠 Basic Usage

  • Submit a custom training job

The following example demonstrates how to submit a custom training job to PAI:

from pai.estimator import Estimator
from pai.image import retrieve

est = Estimator(
    # Retrieve the latest PyTorch image provided by PAI
    image_uri=retrieve(
        framework_name="PyTorch", framework_version="latest"
    ).image_uri,
    command="echo hello",
    # Optionally, specify the source_dir to upload your training code:
    # source_dir="./train_src",
    instance_type="ecs.c6.large",
)

# Submit the training job
est.fit()

print(est.model_data())
  • Deploy Large Language Model

PAI provides numerous pretrained models that you can easily deploy using the PAI Python SDK:

from pai.model import RegisteredModel

# Retrieve the QWen1.5-7b model provided by PAI
qwen_model = RegisteredModel("qwen1.5-7b-chat", model_provider="pai")

# Deploy the model
p = qwen_model.deploy(service_name="qwen_service")

# Call the service
p.predict(
    data={
        "prompt": "How to install PyTorch?",
        "system_prompt": "You are helpful assistant.",
        "temperature": 0.8,
    }
)

# Call the LLM service with openai SDK.
openai_client = p.openai()
res = openai_client.chat.completions.create(
    model="default",
    max_tokens=1024,
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "What is the purpose of life?"}
    ]
)
print(res.choices[0].message.content)
  • Fine-tune the pretrained model

Submit a model fine-tuning task using the fine-tuning script provided by PAI.

from pai.model import ModelTrainingRecipe

# Retrieve the Qwen2-0.5b-instruct model training recipe provided by PAI
training_recipe = ModelTrainingRecipe(
    model_name="qwen2-0.5b-instruct",
    model_provider="pai",
    instance_type="ecs.gn6e-c12g1.3xlarge",
)

# Submit the training job
job = training_recipe.train(
    inputs={
        # Data path on local or Alibaba Cloud OSS (oss://<bucketname>/path/to/data)
        "train": "<YourTrainingDataPath>"
    }
)

# Get output model path
print(training_recipe.model_data())

# Deploy the fine-tuned model
predictor = training_recipe.deploy(service_name="qwen2_finetune")

You can learn more usage examples by visiting the PAI example repository: pai-examples

🤝 Contributing

Contributions to the PAI Python SDK are welcome. Please read our contribution guidelines in the CONTRIBUTING file.

📝 License

PAI Python SDK is developed by Alibaba Cloud and licensed under the Apache License (Version 2.0).

📬 Contact

For support or inquiries, please open an issue on the GitHub repository or contact us in the DingTalk group:

DingTalkGroup