[MRV2] Miscellaneous code cleanup (#57980)
This commit is contained in:
@@ -2,7 +2,6 @@
|
||||
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
||||
from dataclasses import replace
|
||||
from types import SimpleNamespace
|
||||
from unittest.mock import MagicMock
|
||||
|
||||
import numpy as np
|
||||
import pytest
|
||||
@@ -170,38 +169,44 @@ def test_graph_padding_cannot_be_smaller_than_largest_pcp_rank(monkeypatch):
|
||||
)
|
||||
|
||||
|
||||
@pytest.mark.skipif(not torch.cuda.is_available(), reason="needs GPU kernels")
|
||||
@pytest.mark.parametrize(
|
||||
("cg_mode", "num_reqs", "expected_tokens", "expected_reqs"),
|
||||
[
|
||||
(CUDAGraphMode.NONE, 4, None, None),
|
||||
(CUDAGraphMode.PIECEWISE, None, 8, None),
|
||||
(CUDAGraphMode.NONE, 4, 2, 2),
|
||||
(CUDAGraphMode.PIECEWISE, None, 8, 2),
|
||||
(CUDAGraphMode.FULL, 4, 8, 4),
|
||||
],
|
||||
)
|
||||
def test_partition_padding_is_derived_from_batch_descriptor(
|
||||
cg_mode, num_reqs, expected_tokens, expected_reqs
|
||||
):
|
||||
manager = MagicMock()
|
||||
input_batch = MagicMock()
|
||||
manager.partition_batch.return_value = input_batch
|
||||
batch_desc = BatchExecutionDescriptor(
|
||||
cg_mode=cg_mode,
|
||||
num_tokens=8,
|
||||
num_reqs=num_reqs,
|
||||
"""partition_batch derives graph padding from the batch descriptor: no
|
||||
padding without graphs, token padding only for PIECEWISE, and request
|
||||
padding on top for FULL."""
|
||||
device = torch.device("cuda:0")
|
||||
manager = PCPManager(
|
||||
pcp_world_size=2,
|
||||
pcp_rank=0,
|
||||
device=device,
|
||||
max_num_reqs=4,
|
||||
max_num_tokens=8,
|
||||
)
|
||||
global_batch = _make_global_decode_batch(
|
||||
[16, 24], InputBuffers(4, 8, device), device
|
||||
)
|
||||
|
||||
result = pcp_manager_module.maybe_partition_pcp_batch(
|
||||
manager,
|
||||
input_batch,
|
||||
batch_desc,
|
||||
local_batch = manager.partition_batch(
|
||||
global_batch,
|
||||
BatchExecutionDescriptor(
|
||||
cg_mode=cg_mode,
|
||||
num_tokens=8,
|
||||
num_reqs=num_reqs,
|
||||
),
|
||||
)
|
||||
|
||||
assert result is input_batch
|
||||
manager.partition_batch.assert_called_once_with(
|
||||
input_batch,
|
||||
padded_num_tokens=expected_tokens,
|
||||
padded_num_reqs=expected_reqs,
|
||||
)
|
||||
assert local_batch.num_tokens_after_padding == expected_tokens
|
||||
assert local_batch.num_reqs_after_padding == expected_reqs
|
||||
|
||||
|
||||
def test_capture_uses_pcp_persistent_inputs():
|
||||
@@ -472,8 +477,11 @@ def test_partition_defers_dcp_metadata_to_post_partition_batch():
|
||||
|
||||
local_batch = manager.partition_batch(
|
||||
global_batch,
|
||||
padded_num_tokens=4,
|
||||
padded_num_reqs=4,
|
||||
BatchExecutionDescriptor(
|
||||
cg_mode=CUDAGraphMode.FULL,
|
||||
num_tokens=4,
|
||||
num_reqs=4,
|
||||
),
|
||||
)
|
||||
|
||||
assert local_batch.dcp_local_seq_lens is None
|
||||
@@ -489,7 +497,7 @@ def test_partition_defers_dcp_metadata_to_post_partition_batch():
|
||||
|
||||
# What execute_model does next: derive DCP metadata from the final batch
|
||||
# on the PCP-owned buffers.
|
||||
local_batch.dcp_local_seq_lens = gpu_cp_utils.maybe_prepare_dcp_local_seq_lens(
|
||||
local_batch.dcp_local_seq_lens = gpu_cp_utils.prepare_dcp_local_seq_lens(
|
||||
manager.input_buffers.dcp_local_seq_lens,
|
||||
local_batch.seq_lens,
|
||||
local_batch.num_reqs,
|
||||
|
||||
Reference in New Issue
Block a user