Semantix GenAI Serve

Semantix GenAI Serve is a library designed for users of Semantix GenAI Hub to create their own servers running AI models. This library provides an easy-to-use interface for serving AI models using the scalable infrastructure of Semantix GenAI Hub.

Features

• Easy integration with Hugging Face Transformers models
• Support for serving Seq2Seq models
• Customizable server settings
• Built-in support for GPU acceleration

Installation

To install the Semantix GenAI Serve library, run the following command:

pip install semantix-genai-serve

Usage

Basic Example

Here's a simple example of how to use the Semantix GenAI Serve library to serve a Hugging Face Transformers model:

from semantix_genai_serve import SemantixTorchKserve
from semantix_genai_serve.huggingface import ServeAutoSeq2SeqLM

class MyModel(ServeAutoSeq2SeqLM):
    def predict(self, payload, headers):
        # Implement your custom inference logic here
        pass

model = MyModel(checkpoint="facebook/bart-large-cnn")
model.start_server()

Customizing Server Settings

You can customize the server settings by passing additional arguments to the SemantixTorchKserve constructor:

model = MyModel(
    checkpoint="facebook/bart-large-cnn",
    name="my_model",
    base_cache_dir="/path/to/cache",
    force_local_load=True
)

• name: The name of the predictor (default: "predictor")
• base_cache_dir: The base directory for caching models (default: "/mnt/models"). Do not change that when deploying to Semantix GenAI Hub.
• force_local_load: If set to True, the model will be loaded from the local cache directory instead of downloading from Hugging Face's model hub (default: False). Do not change that when deploying to Semantix GenAI Hub.

Implementing Custom Inference Logic

To implement custom inference logic, you need to override the predict method in your model class:

class MyModel(ServeAutoSeq2SeqLM):
    def predict(self, payload, headers):
        # Implement your custom inference logic here
        input_example = payload["input_example"]
        # do something for inference
        # your output must be a dictionary so it gets automatically converted to JSON
        return {"response": "output"}

Contributing

We welcome contributions to the Semantix GenAI Serve library! If you have any suggestions, bug reports, or feature requests, please open an issue on our GitHub repository.

License

Semantix GenAI Serve is released under the MIT License.