o ߥµi®ã@s0ddlmZmZddlmZGdd„deƒZdS)é)ÚABCÚabstractmethod)ÚDictc@sHeZdZdZdd„Zededefdd„ƒZedd „ƒZedd d „ƒZ d S)ÚMetriczÙThe metric base class for computing metrics. The subclasses can either compute a single metric like 'accuracy', or compute the complex metrics for a specific task with or without other Metric subclasses. cOsdS)N©)ÚselfÚargsÚkwargsrrúK/home/ubuntu/.local/lib/python3.10/site-packages/modelscope/metrics/base.pyÚ__init__ szMetric.__init__ÚoutputsÚinputscCódS)a3 Append logits and labels within an eval loop. Will be called after every batch finished to gather the model predictions and the labels. Args: outputs: The model prediction outputs. inputs: The mini batch inputs from the dataloader. Returns: None Nr)rr r rrr Úadds z Metric.addcCr)z²Evaluate the metrics after the eval finished. Will be called after the whole validation finished. Returns: The actual metric dict with standard names. Nr)rrrr Úevaluates zMetric.evaluateÚothercCr)a* When using data parallel, the data required for different metric calculations are stored in their respective Metric classes, and we need to merge these data to uniformly calculate metric. Args: other: Another Metric instance. Returns: None Nr)rrrrr Úmerge*sz Metric.mergeN)rr) Ú__name__Ú __module__Ú __qualname__Ú__doc__r rrrrrrrrr rs  rN)ÚabcrrÚtypingrrrrrr Ús