verl/models/transformers/llama.py

• llama_flash_attn_forward · function · L27-L145 — def llama_flash_attn_forward( self, hidden_states: torch.Tensor, attention_mask: Optional[torch.LongTensor] = None, position_ids: Optional[torch.LongTensor] = None, past_key_value: Optional[Cache] = None, output_attentions: bool = False, use_cache: bool = False, cache_position: Optional[torch.LongTensor] = None, position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # will become mandatory in v4.46 **kwargs, ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]