From 89e1316eae9aae634c174c040415289fe4484aee Mon Sep 17 00:00:00 2001 From: Shuwen Wang <47200617+alphabetc1@users.noreply.github.com> Date: Tue, 29 Sep 2026 07:41:07 +0800 Subject: [PATCH] [mem cache] refactor: remove the index-K continuous getters orphaned by the CP v1 removal (#41469) --- .../sglang/srt/mem_cache/index_key_cache.py | 14 ------- python/sglang/srt/mem_cache/memory_pool.py | 42 ------------------- .../sglang/srt/mem_cache/swa_memory_pool.py | 12 ------ 3 files changed, 68 deletions(-) diff --git a/python/sglang/srt/mem_cache/index_key_cache.py b/python/sglang/srt/mem_cache/index_key_cache.py index 166d5d222d..0975e177a3 100644 --- a/python/sglang/srt/mem_cache/index_key_cache.py +++ b/python/sglang/srt/mem_cache/index_key_cache.py @@ -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, diff --git a/python/sglang/srt/mem_cache/memory_pool.py b/python/sglang/srt/mem_cache/memory_pool.py index 89008bfe21..aa8a3b44bf 100644 --- a/python/sglang/srt/mem_cache/memory_pool.py +++ b/python/sglang/srt/mem_cache/memory_pool.py @@ -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, diff --git a/python/sglang/srt/mem_cache/swa_memory_pool.py b/python/sglang/srt/mem_cache/swa_memory_pool.py index 3ccbbaa55a..93db9fe064 100644 --- a/python/sglang/srt/mem_cache/swa_memory_pool.py +++ b/python/sglang/srt/mem_cache/swa_memory_pool.py @@ -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