Unverified Commit 7f399e4b authored by huangtingwei's avatar huangtingwei Committed by GitHub
Browse files

[HiCacheStorage]support page_first_direct layout for generic set&get (#10522)

parent 873d858b
......@@ -466,6 +466,9 @@ class MHATokenToKVPoolHost(HostKVCache):
return self.kv_buffer[:, :, index : index + self.page_size, :, :].flatten()
elif self.layout == "page_first":
return self.kv_buffer[:, index : index + self.page_size, :, :, :].flatten()
elif self.layout == "page_first_direct":
real_index = index // self.page_size
return self.kv_buffer[:, real_index : real_index + 1, :, :, :, :].flatten()
else:
raise ValueError(f"Unsupported layout: {self.layout}")
......@@ -494,6 +497,13 @@ class MHATokenToKVPoolHost(HostKVCache):
2, self.page_size, self.layer_num, self.head_num, self.head_dim
)
)
elif self.layout == "page_first_direct":
real_index = index // self.page_size
self.kv_buffer[:, real_index : real_index + 1, :, :, :, :] = (
data_page.reshape(
2, 1, self.layer_num, self.page_size, self.head_num, self.head_dim
)
)
else:
raise ValueError(f"Unsupported layout: {self.layout}")
......@@ -731,6 +741,9 @@ class MLATokenToKVPoolHost(HostKVCache):
return self.kv_buffer[:, index : index + self.page_size, :, :].flatten()
elif self.layout == "page_first":
return self.kv_buffer[index : index + self.page_size, :, :, :].flatten()
elif self.layout == "page_first_direct":
real_index = index // self.page_size
return self.kv_buffer[real_index : real_index + 1, :, :, :, :].flatten()
else:
raise ValueError(f"Unsupported layout: {self.layout}")
......@@ -762,6 +775,15 @@ class MLATokenToKVPoolHost(HostKVCache):
1,
self.kv_lora_rank + self.qk_rope_head_dim,
)
elif self.layout == "page_first_direct":
real_index = index // self.page_size
self.kv_buffer[real_index : real_index + 1, :, :, :, :] = data_page.reshape(
1,
self.layer_num,
self.page_size,
1,
self.kv_lora_rank + self.qk_rope_head_dim,
)
else:
raise ValueError(f"Unsupported layout: {self.layout}")
......
......@@ -238,6 +238,19 @@ class TestHiCacheStorageLayerFirstDirectIO(HiCacheStorageBaseMixin, CustomTestCa
return server_args, {}
class TestHiCacheStoragePageFirstDirectIO(HiCacheStorageBaseMixin, CustomTestCase):
"""Page first direct tests for HiCache Storage functionality"""
@classmethod
def _get_additional_server_args_and_env(cls):
"""Get additional server arguments specific to configuration - override in subclasses"""
server_args = {
"--hicache-mem-layout": "page_first_direct",
"--hicache-io-backend": "direct",
}
return server_args, {}
class TestHiCacheStoragePageFirstLayout(HiCacheStorageBaseMixin, CustomTestCase):
"""Page first layout tests for HiCache Storage functionality"""
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment