mistralrs_core/dummy_paged_attention/layers/
paged_attention.rsuse candle_core::{Device, Result, Tensor};
use crate::{
attention::SdpaParams,
pipeline::text_models_inputs_processor::{FlashParams, PagedAttentionInputMetadata},
};
#[allow(dead_code)]
pub struct PagedAttention;
impl PagedAttention {
pub fn new(
_head_dim: usize,
_device: &Device,
_alibi_slopes: Option<Vec<f32>>,
) -> Result<Self> {
unreachable!();
}
#[allow(clippy::too_many_arguments)]
#[allow(unused_variables)]
pub fn forward(
&self,
_query: &Tensor,
_key: &Tensor,
_value: &Tensor,
_attention_mask: Option<&Tensor>,
_key_cache: Option<Tensor>,
_value_cache: Option<Tensor>,
_input_metadata: &PagedAttentionInputMetadata,
_sdpa_params: &SdpaParams,
_flash_params: Option<&FlashParams>,
) -> Result<Tensor> {
unreachable!();
}
}