我正在使用Huggingface Trainer API来微调ASR模型,例如。
https://huggingface.co/openai/whisper-tiny
在回调函数中,我调用evaluate API来计算
CER metric
.
{{code-snippet-needed}} # i.e. What have you tried?
@ramraj-chandradevan, please add your code snippet here to show what you have tried.
它为每个验证步骤输出一个CER分数,例如。
@ramraj-chandradevan, please add an example of what numbers you see at each validation step
然而,在我的案例中,我试图获得每个验证示例的CER分数,而不是整个样本集的一个值。
@ramraj-chandradevan, please add an example of the expected behavior you'll like to see during the validation step. What numbers do you see at each validation step instead of the default single value?
如果我调用每个示例的度量,则速度非常慢,例如。
{{code-snippet-needed}} # i.e. What have you tried when calling CER score again to get validation numbers for each example?
@ramraj-chandradevan, please add code snippet that you've tried when changing the `compute_metrics` functions that you're reporting as "very slow"
有人知道在模型验证过程中以任何方式将获得CER分数的预期行为输出到每个示例吗?
