[mem cache] refactor: remove the index-K continuous getters orphaned by the CP v1 removal (#41469)
This commit is contained in:
@@ -65,20 +65,6 @@ class IndexKeyCache:
|
||||
def get_buffer(self, layer_id: int) -> torch.Tensor:
|
||||
return self.get_local_buffer(layer_id)
|
||||
|
||||
def get_k_continuous(self, layer_id: int, seq_len: int, page_indices: torch.Tensor):
|
||||
buf = self.get_buffer(layer_id)
|
||||
return index_buf_accessor.GetK.execute(
|
||||
self.pool, buf, seq_len=seq_len, page_indices=page_indices
|
||||
)
|
||||
|
||||
def get_k_scale_continuous(
|
||||
self, layer_id: int, seq_len: int, page_indices: torch.Tensor
|
||||
):
|
||||
buf = self.get_buffer(layer_id)
|
||||
return index_buf_accessor.GetS.execute(
|
||||
self.pool, buf, seq_len=seq_len, page_indices=page_indices
|
||||
)
|
||||
|
||||
def get_k_and_scale(
|
||||
self,
|
||||
layer_id: int,
|
||||
|
||||
@@ -4317,30 +4317,6 @@ class HybridLinearKVPool(KVCache):
|
||||
layer_id = self._transfer_full_attention_id(layer_id)
|
||||
self.full_kv_pool.invalidate_index_buffer_for_layer(layer_id)
|
||||
|
||||
def get_index_k_continuous(
|
||||
self,
|
||||
layer_id: int,
|
||||
seq_len: int,
|
||||
page_indices: torch.Tensor,
|
||||
):
|
||||
assert self.use_dsa, "get_index_k_continuous called when use_dsa is False"
|
||||
self._wait_for_layer(layer_id)
|
||||
layer_id = self._transfer_full_attention_id(layer_id)
|
||||
return self.full_kv_pool.get_index_k_continuous(layer_id, seq_len, page_indices)
|
||||
|
||||
def get_index_k_scale_continuous(
|
||||
self,
|
||||
layer_id: int,
|
||||
seq_len: int,
|
||||
page_indices: torch.Tensor,
|
||||
):
|
||||
assert self.use_dsa, "get_index_k_scale_continuous called when use_dsa is False"
|
||||
self._wait_for_layer(layer_id)
|
||||
layer_id = self._transfer_full_attention_id(layer_id)
|
||||
return self.full_kv_pool.get_index_k_scale_continuous(
|
||||
layer_id, seq_len, page_indices
|
||||
)
|
||||
|
||||
def get_index_k_scale_buffer(
|
||||
self,
|
||||
layer_id: int,
|
||||
@@ -5146,24 +5122,6 @@ class DSATokenToKVPool(MLATokenToKVPool):
|
||||
def get_index_k_with_scale_buffer(self, layer_id: int) -> torch.Tensor:
|
||||
return self.index_key_cache.get_local_buffer(layer_id)
|
||||
|
||||
def get_index_k_continuous(
|
||||
self,
|
||||
layer_id: int,
|
||||
seq_len: int,
|
||||
page_indices: torch.Tensor,
|
||||
):
|
||||
return self.index_key_cache.get_k_continuous(layer_id, seq_len, page_indices)
|
||||
|
||||
def get_index_k_scale_continuous(
|
||||
self,
|
||||
layer_id: int,
|
||||
seq_len: int,
|
||||
page_indices: torch.Tensor,
|
||||
):
|
||||
return self.index_key_cache.get_k_scale_continuous(
|
||||
layer_id, seq_len, page_indices
|
||||
)
|
||||
|
||||
def get_index_k_scale_buffer(
|
||||
self,
|
||||
layer_id: int,
|
||||
|
||||
@@ -305,18 +305,6 @@ class SWAKVPool(BaseSWAKVPool):
|
||||
assert not is_swa_layer
|
||||
return self.full_kv_pool.get_index_k_with_scale_buffer(layer_id_pool)
|
||||
|
||||
def get_index_k_continuous(self, layer_id: int, *args, **kwargs):
|
||||
layer_id_pool, is_swa_layer = self.layers_mapping[layer_id]
|
||||
assert not is_swa_layer
|
||||
return self.full_kv_pool.get_index_k_continuous(layer_id_pool, *args, **kwargs)
|
||||
|
||||
def get_index_k_scale_continuous(self, layer_id: int, *args, **kwargs):
|
||||
layer_id_pool, is_swa_layer = self.layers_mapping[layer_id]
|
||||
assert not is_swa_layer
|
||||
return self.full_kv_pool.get_index_k_scale_continuous(
|
||||
layer_id_pool, *args, **kwargs
|
||||
)
|
||||
|
||||
def get_index_k_scale_buffer(self, layer_id: int, *args, **kwargs):
|
||||
layer_id_pool, is_swa_layer = self.layers_mapping[layer_id]
|
||||
assert not is_swa_layer
|
||||
|
||||
Reference in New Issue
Block a user