:class:`transformers.LogitsProcessor` that enforces that when specified sequences appear, stop geration. Args: stop_words_ids (:

Integer values for each supported action type in AndroidInTheWild.

Trainer for the Group Relative Policy Optimization (GRPO) method. This algorithm was initially proposed in the paper [DeepSeekMath: Pushing t

管理全局任务同步信号并分配优势值。

平衡本地机器创建的数据和任务，并与全局同步。

A 2D perceiver-resampler network with one cross attention layers by (grid_size**2) learnable queries and 2d sincos pos_emb Outputs

self-attention layer class. Self-attention layer takes input with size [s, b, h] and returns output of the same size.

Script arguments for the GRPO training script. Args: reward_funcs (`list[str]`): List of reward functions. Possible valu

r""" Configuration class for the [`GRPOTrainer`]. Only the parameters specific to GRPO training are listed here. For details on other paramet

Multiturn RFT Dataset

QWen tokenizer.

Dataset for supervised fine-tuning.