@smart-rank-model

Convenient interface for utilizing various ranking models via API and locally.

Features

• Supports multiple ranking models
• Flexible adapter system for different model implementations
• GPU acceleration support (when available)
• Easy-to-use API for document ranking

Configuration

The SmartRankModel constructor accepts two parameters:

1. env: The environment object containing adapter configurations
2. opts: Model configuration options

Model Options

• model_key: Identifier for the model in the models.json file
• adapter: The adapter to use for this model
• use_gpu: Boolean to enable/disable GPU acceleration (auto-detected if not specified)
• gpu_batch_size: Batch size for GPU processing (default: 10)

Adapters

Adapters should be implemented and added to the env.opts.smart_rank_adapters object. Each adapter should implement the following methods:

• constructor(model): Initialize the adapter
• load(): Load the model
• rank(query, documents): Rank the documents based on the query

License

MIT License. See LICENSE file for details.