[Misc] Small perf improvements (#6520)
This commit is contained in:
@@ -249,10 +249,13 @@ def test_append_slots(block_size, prompt_len, num_slots_to_append,
|
||||
|
||||
# Expect consumed blocks to be new blocks required to support the new slots.
|
||||
expected_consumed_blocks = len(
|
||||
chunk_list(
|
||||
list(
|
||||
range(prompt_len + num_slots_to_append + num_lookahead_slots)),
|
||||
block_size)) - len(chunk_list(list(range(prompt_len)), block_size))
|
||||
list(
|
||||
chunk_list(
|
||||
list(
|
||||
range(prompt_len + num_slots_to_append +
|
||||
num_lookahead_slots)),
|
||||
block_size))) - len(
|
||||
list(chunk_list(list(range(prompt_len)), block_size)))
|
||||
assert num_consumed_blocks == expected_consumed_blocks
|
||||
|
||||
|
||||
|
||||
@@ -58,10 +58,10 @@ def test_allocate_immutable_block(num_cpu_blocks: int, num_gpu_blocks: int,
|
||||
|
||||
unique_token_ids = list(
|
||||
range((num_cpu_blocks + num_gpu_blocks) * block_size))
|
||||
gpu_token_ids = chunk_list(unique_token_ids[:num_gpu_blocks * block_size],
|
||||
block_size)
|
||||
cpu_token_ids = chunk_list(unique_token_ids[num_gpu_blocks * block_size:],
|
||||
block_size)
|
||||
gpu_token_ids = list(
|
||||
chunk_list(unique_token_ids[:num_gpu_blocks * block_size], block_size))
|
||||
cpu_token_ids = list(
|
||||
chunk_list(unique_token_ids[num_gpu_blocks * block_size:], block_size))
|
||||
|
||||
assert allocator.get_num_free_blocks(Device.CPU) == num_cpu_blocks
|
||||
assert allocator.get_num_free_blocks(Device.GPU) == num_gpu_blocks
|
||||
|
||||
Reference in New Issue
Block a user