Method forward

model/modeling_gemma2.py:451–506

(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_value: Optional[Cache] = None,
        output_attentions: Optional[bool] = False,
        use_cache: Optional[bool] = False,
        cache_position: Optional[torch.LongTensor] = None,
    )

        self.sliding_window = config.sliding_window

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_value: Optional[Cache] = None,
        output_attentions: Optional[bool] = False,
        use_cache: Optional[bool] = False,
        cache_position: Optional[torch.LongTensor] = None,
    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
        if self.is_sliding and attention_mask is not None:  # efficient SDPA and no padding
            # Flash-attn is a 2D tensor
            if self.config._attn_implementation == "flash_attention_2":
                if past_key_value is not None:  # when decoding
                    attention_mask = attention_mask[:, -self.sliding_window :]
            else:
                min_dtype = torch.finfo(hidden_states.dtype).min
                sliding_window_mask = torch.tril(
                    torch.ones_like(attention_mask, dtype=torch.bool), diagonal=-self.sliding_window
                )
                attention_mask = torch.where(sliding_window_mask, min_dtype, attention_mask)
                if attention_mask.shape[-1] <= 1:  # when decoding
                    attention_mask = attention_mask[:, :, :, -self.sliding_window :]

        residual = hidden_states

        hidden_states = self.input_layernorm(hidden_states)

        # Self Attention
        hidden_states, self_attn_weights, present_key_value = self.self_attn(
            hidden_states=hidden_states,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_value=past_key_value,
            output_attentions=output_attentions,
            use_cache=use_cache,
            cache_position=cache_position,
        )
        hidden_states = self.post_attention_layernorm(hidden_states)
        hidden_states = residual + hidden_states

        residual = hidden_states
        hidden_states = self.pre_feedforward_layernorm(hidden_states)
        hidden_states = self.mlp(hidden_states)
        hidden_states = self.post_feedforward_layernorm(hidden_states)
        hidden_states = residual + hidden_states

        outputs = (hidden_states,)

        if output_attentions:
            outputs += (self_attn_weights,)

        if use_cache:
            outputs += (present_key_value,)

        return outputs