[v1] Hybrid Memory Allocator (#17996)

Signed-off-by: Chen Zhang <zhangch99@outlook.com>
This commit is contained in:
Chen Zhang
2025-06-06 11:47:09 +08:00
committed by GitHub
parent 3465b87ef8
commit f8a1a2d108
21 changed files with 1605 additions and 440 deletions

View File

@@ -97,7 +97,7 @@ def create_scheduler(
)
kv_cache_config = KVCacheConfig(
num_blocks=num_blocks, # A large number of blocks to hold all requests
tensors={},
kv_cache_tensors=[],
kv_cache_groups=[
KVCacheGroupSpec(['layer'],
FullAttentionSpec(block_size, 1, 1, torch.float32,
@@ -814,10 +814,10 @@ def _assert_right_kv_cache_manager(
# Make sure the request stats are right.
EXPECTED_TOTAL_BLOCKS = num_tokens // block_size
for req_id in req_ids:
blocks = (scheduler.kv_cache_manager.single_type_manager.
req_to_blocks[req_id])
blocks = (scheduler.kv_cache_manager.coordinator.
single_type_managers[0].req_to_blocks[req_id])
hashes = scheduler.kv_cache_manager.req_to_block_hashes[req_id]
assert (scheduler.kv_cache_manager.single_type_manager.
assert (scheduler.kv_cache_manager.coordinator.single_type_managers[0].
num_cached_block[req_id] == EXPECTED_TOTAL_BLOCKS)
assert len(blocks) == EXPECTED_TOTAL_BLOCKS
assert len(hashes) == EXPECTED_TOTAL_BLOCKS
@@ -1198,11 +1198,11 @@ def assert_scheduler_empty(scheduler: Scheduler):
assert len(scheduler.encoder_cache_manager.cached) == 0
# KVCache Manager.
assert len(
scheduler.kv_cache_manager.single_type_manager.req_to_blocks) == 0
assert len(scheduler.kv_cache_manager.coordinator.single_type_managers[0].
req_to_blocks) == 0
assert len(scheduler.kv_cache_manager.req_to_block_hashes) == 0
assert len(
scheduler.kv_cache_manager.single_type_manager.num_cached_block) == 0
assert len(scheduler.kv_cache_manager.coordinator.single_type_managers[0].
num_cached_block) == 0
num_free_blocks = (
scheduler.kv_cache_manager.block_pool.free_block_queue.num_free_blocks)
assert num_free_blocks == (