samples_per_cls : list[int] # 각 클래스로부터의 원본 샘플 개수 beta : float # 0.9~0.999 사이 권장 (0이면 일반 CE와 동일)

Override to include images/videos in generation for multimodal models.

Identity collator that passes data through unchanged.

Reward function that checks if the completion is correct using either symbolic verification or exact string matching.

Setup the optimizer. We provide a reasonable default that works well. If you want to use something else, you can pass a tuple in the

Setup the optimizer. We provide a reasonable default that works well. If you want to use something else, you can pass a tuple in the

Setup the optimizer. We provide a reasonable default that works well. If you want to use something else, you can pass a tuple in the

Setup the optimizer. We provide a reasonable default that works well. If you want to use something else, you can pass a tuple in the

Override evaluation_loop to support generation-based evaluation. If compute_metrics is provided and prediction_loss_only is False,

Reward function that checks if the completion has a specific format.

logits : [B, C] targets : [B] (long)

logits: [B, C] targets: [B] (long)

Returns the training [`~torch.utils.data.DataLoader`]. Will use no sampler if `train_dataset` does not implement `__len__`, a random

r""" image_grid_thw (`torch.LongTensor` of shape `(num_images, 3)`, *optional*): The temporal, height and width of feature shape of each i

r""" image_grid_thw (`torch.LongTensor` of shape `(num_images, 3)`, *optional*): The temporal, height and width of feature shape of each i

r""" image_grid_thw (`torch.LongTensor` of shape `(num_images, 3)`, *optional*): The temporal, height and width of feature shape of each i