evaluate/main/en/package_reference/hub_methods.md

Hub methods

Methods for using the Hugging Face Hub:

Push to hub [[evaluate.push_to_hub]]

evaluate.push_to_hub[[evaluate.push_to_hub]]

Source

Pushes the result of a metric to the metadata of a model repository in the Hub.

Example:

>>> push_to_hub(
...     model_id="huggingface/gpt2-wikitext2",
...     metric_value=0.5
...     metric_type="bleu",
...     metric_name="BLEU",
...     dataset_name="WikiText",
...     dataset_type="wikitext",
...     dataset_split="test",
...     task_type="text-generation",
...     task_name="Text Generation"
... )

Parameters:

model_id (str) : Model id from https://hf.co/models.

task_type (str) : Task id, refer to the Hub allowed tasks for allowed values.

dataset_type (str) : Dataset id from https://hf.co/datasets.

dataset_name (str) : Pretty name for the dataset.

metric_type (str) : Metric id from https://hf.co/metrics.

metric_name (str) : Pretty name for the metric.

metric_value (float) : Computed metric value.

task_name (str, optional) : Pretty name for the task.

dataset_config (str, optional) : Dataset configuration used in load_dataset. See load_dataset for more info.

dataset_split (str, optional) : Name of split used for metric computation.

dataset_revision (str, optional) : Git hash for the specific version of the dataset.

dataset_args (dict[str, int], optional) : Additional arguments passed to load_dataset.

metric_config (str, optional) : Configuration for the metric (e.g. the GLUE metric has a configuration for each subset).

metric_args (dict[str, int], optional) : Arguments passed during compute().

overwrite (bool, optional, defaults to False) : If set to True an existing metric field can be overwritten, otherwise attempting to overwrite any existing fields will cause an error.