Model Configuration

Configuring an NLP model source can be done by specifying a mapping in the RIME JSON configuration file, under the model_info argument. Models can be defined via a custom model.py file or through a native integration with Hugging Face.

Template

{
    "model_info": {
        "path": "/path/to/model.py"        (REQUIRED)
    }
    ...
}

Arguments

path: string, required

Path to Python model file. For instructions on how to create this Python model file, please see Specify a Model.

Hugging Face Model

Note: this is only supported for the Text Classification and Natural Language Inference tasks.

{
    "model_info": {
        "type": "huggingface_classification",      (REQUIRED)
        "model_uri": "path",                       (REQUIRED)
        "tokenizer_uri": null,
        "model_max_length": null,

    },
    ...
}

Arguments

model_uri: string, required

The pretrained model name or path used to load a pretrained Hugging Face model from disk or from the model hub.
tokenizer_uri: string or null, default = null

The pretrained tokenizer name or path used to load the tokenizer from disk or from the model hub. If null, RIME defaults to loading from the provided model_uri.
model_max_length: int or null, default = null

The maximum sequence length (in tokens) supported by the model. If null, RIME infers the maximum length from the pretrained model and tokenizer.