Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`Phi3DecoderLayer`] Args: config: Phi3Config

Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`Phi3DecoderLayer`] Args: config: Phi3Config

JourneyDB WebDataset dataset.

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

2D Image to Patch Embedding

2D Image to Patch Embedding

Sampler that repeats the indices of a dataset in a structured manner. Args: data_source (`Sized`): Dataset to sample fro

Embeds scalar timesteps into vector representations.

Embeds scalar timesteps into vector representations.

JourneyDB WebDataset dataset.

The final layer of DiT.

The final layer of DiT.

Params for coco evaluation api

Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`Phi3DecoderLayer`] We only modified the attention mask

Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`Phi3DecoderLayer`] We only modified the attention mask

Sampler that repeats the indices of a dataset N times. Args: data_source (`Sized`): Dataset to sample from. repe

args for callbacks, benchmarks etc

Script arguments for the GRPO training script.

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

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

Script arguments for the GRPO training script.

Diffusion model with a Transformer backbone.

Diffusion model with a Transformer backbone.

Multi-headed attention from 'Attention Is All You Need' paper

Multi-headed attention from 'Attention Is All You Need' paper

Phi-3 flash attention module. This module inherits from `Phi3Attention` as the weights of the module stays untouched. The only required chang

Phi-3 flash attention module. This module inherits from `Phi3Attention` as the weights of the module stays untouched. The only required chang

Phi3 attention module using torch.nn.functional.scaled_dot_product_attention. This module inherits from `Phi3Attention` as the weights of the

Phi3 attention module using torch.nn.functional.scaled_dot_product_attention. This module inherits from `Phi3Attention` as the weights of the

args for callbacks, benchmarks etc

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