fea197ad50
### What this PR does / why we need it?
1. ✅ Upgrade vllm commit to: 0115
(8471b27df97c3eb79f891802fc0e858f8f7ac6a0)
Modify import paths due to the refactors：
https://github.com/vllm-project/vllm/pull/32245
https://github.com/vllm-project/vllm/pull/32060
Test result:
https://github.com/vllm-project/vllm-ascend/actions/runs/21034239336/job/60490156965?pr=5913
2. ✅Upgrade vllm commit to: 0119
(9a1f16da1e423ede2c2f52a9850cbfbb39cefe96)
Fix `WorkerProc.__init__() missing 1 required positional argument:
'is_driver_worker'` due to
https://github.com/vllm-project/vllm/pull/28506
Test result:
https://github.com/vllm-project/vllm-ascend/actions/runs/21156263050/job/60841668755?5569
3. ✅Upgrade vllm commit to:
0120(148117ea2e689cd43df4be6892671a17cdae5833)
1. Add `skip_compiled` param in `set_forward_context` due to
https://github.com/vllm-project/vllm/pull/30385
2. Modify `tests/ut/spec_decode/test_eagle_proposer.py` due to
https://github.com/vllm-project/vllm/pull/24322
change `self.max_num_tokens =
vllm_config.scheduler_config.max_num_batched_tokens + max_batch_size`
3. Modify UT import paths due to the
refactors：https://github.com/vllm-project/vllm/pull/32060
Test result:
https://github.com/vllm-project/vllm-ascend/actions/runs/21204851770/job/60999046946
4. ✅Upgrade vllm commit to:
0121(f23fb5a7c1b61350c5c40ca1115d3bf8cf2b8cc9)
1. vLLM switched `uses_mrope` from target to draft model config, making
`positions`/`mrope_positions` mutually exclusive, breaking vllm-ascend's
direct self.positions access and tests missing
`draft_model_config.uses_mrope`.
https://github.com/vllm-project/vllm/pull/32048
2. Moved bs_to_padded_graph_size from CompilationConfig to
CudagraphDispatcher due to the refactor
https://github.com/vllm-project/vllm/pull/30143
3. Remove unused `maybe_setup_kv_connector` due to
https://github.com/vllm-project/vllm/pull/32077
Test result:
https://github.com/vllm-project/vllm-ascend/actions/runs/21217728738/job/61043738834
6. ✅Upgrade vllm commit to:
0122(8ebf271bb6d1e7e9b1a55be73d755ef1a57dbbe5)
Updating FusedMoEParallelConfig (added enable_eplb) and FusedMoEConfig
due to https://github.com/vllm-project/vllm/pull/32414
Test result:
https://github.com/vllm-project/vllm-ascend/actions/runs/21249922546/job/61148613054
8. ✅Upgrade vllm commit to:
0123(dc917cceb877dfd13f98c538c4c96158047d98bd)
Setting temperature=0.0 due to the removal of the default temperature
value in https://github.com/vllm-project/vllm/pull/32723
Test result:
https://github.com/vllm-project/vllm-ascend/actions/runs/21280796875
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.14.0
- vLLM main:
d68209402d
---------
Signed-off-by: wjunLu <wjunlu217@gmail.com>
Signed-off-by: Meihan-chen <jcccx.cmh@gmail.com>
Co-authored-by: wjunLu <wjunlu217@gmail.com>
280 lines
12 KiB
Python
280 lines
12 KiB
Python
import sys
|
|
from unittest.mock import MagicMock, patch
|
|
|
|
import torch
|
|
|
|
from tests.ut.base import TestBase
|
|
from vllm_ascend.attention.attention_v1 import AscendAttentionState
|
|
from vllm.distributed.parallel_state import GroupCoordinator
|
|
|
|
if 'torch_npu._inductor' not in sys.modules:
|
|
sys.modules['torch_npu._inductor'] = MagicMock()
|
|
|
|
from vllm_ascend.attention.sfa_v1 import (AscendSFABackend, AscendSFAImpl,
|
|
AscendSFAMetadata,
|
|
AscendSFAMetadataBuilder)
|
|
from vllm_ascend.utils import enable_dsa_cp, vllm_version_is
|
|
|
|
|
|
class TestAscendSFABackend(TestBase):
|
|
|
|
def test_get_name(self):
|
|
self.assertEqual(AscendSFABackend.get_name(), "ASCEND_SFA")
|
|
|
|
def test_get_builder_cls(self):
|
|
self.assertEqual(AscendSFABackend.get_builder_cls(),
|
|
AscendSFAMetadataBuilder)
|
|
|
|
def test_get_kv_cache_shape(self):
|
|
result = AscendSFABackend.get_kv_cache_shape(2, 4, 8, 128)
|
|
self.assertEqual(result, (2, 4, 8, 128))
|
|
|
|
def test_get_impl_cls(self):
|
|
result = AscendSFABackend.get_impl_cls()
|
|
self.assertEqual(result, AscendSFAImpl)
|
|
|
|
|
|
class TestAscendSFAMetadata(TestBase):
|
|
|
|
def test_ascend_sfa_metadata_default(self):
|
|
num_actual_tokens = 100
|
|
slot_mapping = torch.randn(100, 4, 1024)
|
|
seq_lens = torch.tensor([30, 50])
|
|
cum_query_lens = torch.tensor([0, 30, 80])
|
|
block_tables = torch.randint(0, 100, (100, 4))
|
|
|
|
rope_dim = 32
|
|
max_seq_len = int(seq_lens.max().item())
|
|
sin = torch.randn(max_seq_len, rope_dim)
|
|
cos = torch.randn(max_seq_len, rope_dim)
|
|
|
|
num_input_tokens = 2
|
|
head_dim = None
|
|
attn_mask = None
|
|
attn_state = AscendAttentionState.ChunkedPrefill
|
|
|
|
metadata = AscendSFAMetadata(
|
|
num_actual_tokens=num_actual_tokens,
|
|
slot_mapping=slot_mapping,
|
|
seq_lens=seq_lens,
|
|
cum_query_lens=cum_query_lens,
|
|
block_tables=block_tables,
|
|
sin=sin,
|
|
cos=cos,
|
|
num_input_tokens=num_input_tokens,
|
|
head_dim=head_dim,
|
|
attn_mask=attn_mask,
|
|
attn_state=attn_state,
|
|
)
|
|
|
|
self.assertEqual(metadata.num_actual_tokens, num_actual_tokens)
|
|
self.assertIs(metadata.slot_mapping, slot_mapping)
|
|
self.assertTrue(torch.equal(metadata.seq_lens, seq_lens))
|
|
self.assertTrue(torch.equal(metadata.cum_query_lens, cum_query_lens))
|
|
self.assertIs(metadata.block_tables, block_tables)
|
|
self.assertIs(metadata.sin, sin)
|
|
self.assertIs(metadata.cos, cos)
|
|
self.assertEqual(metadata.num_input_tokens, num_input_tokens)
|
|
self.assertIs(metadata.head_dim, head_dim)
|
|
self.assertIs(metadata.attn_mask, attn_mask)
|
|
self.assertEqual(metadata.attn_state, attn_state)
|
|
|
|
|
|
class TestAscendSFAMetadataBuilder(TestBase):
|
|
|
|
@patch('vllm.distributed.parallel_state._TP',
|
|
new_callable=lambda: MagicMock(spec=GroupCoordinator))
|
|
def setUp(self, mock_tp):
|
|
mock_tp.world_size = 2
|
|
mock_tp.rank_in_group = MagicMock()
|
|
mock_tp.device_group = MagicMock()
|
|
|
|
self.mock_cfg = MagicMock()
|
|
|
|
self.mock_cfg.parallel_config = MagicMock()
|
|
self.mock_cfg.parallel_config.tensor_parallel_size = 1
|
|
self.mock_cfg.parallel_config.prefill_context_parallel_size = 1
|
|
self.mock_cfg.parallel_config.decode_context_parallel_size = 1
|
|
|
|
self.mock_cfg.compilation_config = MagicMock()
|
|
self.mock_cfg.compilation_config.pass_config = MagicMock()
|
|
self.mock_cfg.compilation_config.pass_config.enable_sp = False
|
|
|
|
self.mock_cfg.speculative_config.num_speculative_tokens = 0
|
|
|
|
self.patcher = patch("vllm.config.get_current_vllm_config",
|
|
return_value=self.mock_cfg)
|
|
self.patcher.start()
|
|
|
|
# Mock parent class __init__ to avoid complex initialization,
|
|
# but still set the essential attributes that child class needs
|
|
def mock_parent_init(self, kv_cache_spec, layer_names, vllm_config,
|
|
device, metadata_cls, supports_dcp_with_varlen):
|
|
self.metadata_cls = metadata_cls
|
|
self.kv_cache_spec = kv_cache_spec
|
|
self.model_config = vllm_config.model_config
|
|
self.vllm_config = vllm_config
|
|
self.device = device
|
|
self.chunked_prefill_workspace_size = 128 * 1024
|
|
self.chunked_prefill_workspace = torch.empty(
|
|
(self.chunked_prefill_workspace_size,
|
|
vllm_config.model_config.get_head_size()),
|
|
dtype=vllm_config.model_config.dtype,
|
|
device=device,
|
|
)
|
|
|
|
self.parent_init_patcher = patch(
|
|
("vllm.v1.attention.backends.mla.common.MLACommonMetadataBuilder.__init__"
|
|
if vllm_version_is('0.14.1') else
|
|
"vllm.model_executor.layers.attention.mla_attention.MLACommonMetadataBuilder.__init__"),
|
|
mock_parent_init)
|
|
self.parent_init_patcher.start()
|
|
|
|
if hasattr(enable_dsa_cp, "cache_clear"):
|
|
enable_dsa_cp.cache_clear()
|
|
|
|
def tearDown(self):
|
|
self.patcher.stop()
|
|
self.parent_init_patcher.stop()
|
|
|
|
def test_ascend_sfa_metadata_builder_default(self):
|
|
kv_cache_spec = MagicMock()
|
|
layer_names = ["layer1", "layer2"]
|
|
vllm_config = MagicMock()
|
|
vllm_config.cache_config.block_size = 16
|
|
vllm_config.model_config.max_model_len = 1024
|
|
vllm_config.model_config.get_head_size.return_value = 64
|
|
vllm_config.model_config.dtype = torch.float16
|
|
vllm_config.model_config.hf_text_config.qk_rope_head_dim = 64
|
|
speculative_config = MagicMock()
|
|
speculative_config.num_speculative_tokens = 4
|
|
vllm_config.speculative_config = speculative_config
|
|
device = torch.device("cpu")
|
|
|
|
builder = AscendSFAMetadataBuilder(kv_cache_spec=kv_cache_spec,
|
|
layer_names=layer_names,
|
|
vllm_config=vllm_config,
|
|
device=device)
|
|
|
|
assert builder.device == device
|
|
assert builder.vllm_config == vllm_config
|
|
|
|
@patch("vllm_ascend.attention.sfa_v1.get_current_vllm_config")
|
|
@patch("vllm_ascend.attention.sfa_v1.get_cos_and_sin_mla")
|
|
@patch("vllm_ascend.attention.sfa_v1.enable_dsa_cp")
|
|
def test_ascend_sfa_metadata_builder_build(
|
|
self,
|
|
mock_enable_dsa_cp,
|
|
mock_get_cos_and_sin_mla,
|
|
mock_get_current_vllm_config,
|
|
):
|
|
mock_enable_dsa_cp.return_value = False
|
|
|
|
cfg = MagicMock()
|
|
cfg.model_config = MagicMock()
|
|
cfg.model_config.hf_text_config = MagicMock()
|
|
|
|
mock_get_current_vllm_config.return_value = cfg
|
|
kv_cache_spec = MagicMock()
|
|
layer_names = ["layer1", "layer2"]
|
|
vllm_config = MagicMock()
|
|
vllm_config.cache_config.block_size = 16
|
|
vllm_config.model_config.max_model_len = 1024
|
|
vllm_config.model_config.get_head_size.return_value = 64
|
|
vllm_config.model_config.dtype = torch.float16
|
|
vllm_config.model_config.hf_text_config.qk_rope_head_dim = 64
|
|
speculative_config = MagicMock()
|
|
speculative_config.num_speculative_tokens = 4
|
|
vllm_config.speculative_config = speculative_config
|
|
device = torch.device("cpu")
|
|
|
|
builder = AscendSFAMetadataBuilder(kv_cache_spec=kv_cache_spec,
|
|
layer_names=layer_names,
|
|
vllm_config=vllm_config,
|
|
device=device)
|
|
|
|
common_attn_metadata = MagicMock()
|
|
common_attn_metadata.num_reqs = 10
|
|
common_attn_metadata.num_actual_tokens = 100
|
|
common_attn_metadata.query_start_loc = torch.tensor(
|
|
[0, 10, 20, 30, 40, 50, 60, 70, 80, 90])
|
|
common_attn_metadata.query_start_loc_cpu = torch.tensor(
|
|
[0, 10, 20, 30, 40, 50, 60, 70, 80, 90])
|
|
common_attn_metadata.slot_mapping = torch.randn(100, 4, 1024)
|
|
common_attn_metadata.seq_lens_cpu = torch.tensor([2] * 10)
|
|
common_attn_metadata.positions = torch.randn(100)
|
|
common_attn_metadata.attn_mask = None
|
|
common_attn_metadata.attn_state = AscendAttentionState.ChunkedPrefill
|
|
common_attn_metadata.block_table_tensor = torch.randn(100, 4)
|
|
common_attn_metadata.cos = None
|
|
common_attn_metadata.sin = None
|
|
common_attn_metadata.num_input_tokens = 100
|
|
|
|
mock_get_cos_and_sin_mla.return_value = (torch.randn(100),
|
|
torch.randn(100))
|
|
|
|
metadata = builder.build(
|
|
common_prefix_len=10,
|
|
common_attn_metadata=common_attn_metadata,
|
|
)
|
|
|
|
assert isinstance(metadata, AscendSFAMetadata)
|
|
assert metadata.num_actual_tokens == common_attn_metadata.num_actual_tokens
|
|
assert metadata.slot_mapping.shape == (100, 4, 1024)
|
|
|
|
@patch("vllm_ascend.attention.sfa_v1.get_current_vllm_config")
|
|
@patch("vllm_ascend.attention.sfa_v1.get_cos_and_sin_mla")
|
|
def test_ascend_sfa_metadata_builder_build_for_graph_capture(
|
|
self, mock_get_cos_and_sin_mla, mock_get_current_vllm_config):
|
|
cfg = MagicMock()
|
|
cfg.model_config = MagicMock()
|
|
cfg.model_config.hf_text_config = MagicMock()
|
|
|
|
mock_get_current_vllm_config.return_value = cfg
|
|
|
|
kv_cache_spec = MagicMock()
|
|
layer_names = ["layer1", "layer2"]
|
|
vllm_config = MagicMock()
|
|
vllm_config.cache_config.block_size = 16
|
|
vllm_config.model_config.max_model_len = 1024
|
|
vllm_config.model_config.get_head_size.return_value = 64
|
|
vllm_config.model_config.dtype = torch.float16
|
|
vllm_config.model_config.hf_text_config.qk_rope_head_dim = 64
|
|
speculative_config = MagicMock()
|
|
speculative_config.num_speculative_tokens = 4
|
|
vllm_config.speculative_config = speculative_config
|
|
device = torch.device("cpu")
|
|
|
|
builder = AscendSFAMetadataBuilder(kv_cache_spec=kv_cache_spec,
|
|
layer_names=layer_names,
|
|
vllm_config=vllm_config,
|
|
device=device)
|
|
|
|
common_attn_metadata = MagicMock()
|
|
common_attn_metadata.num_reqs = 10
|
|
common_attn_metadata.num_actual_tokens = 100
|
|
common_attn_metadata.query_start_loc = torch.tensor(
|
|
[0, 10, 20, 30, 40, 50, 60, 70, 80, 90])
|
|
common_attn_metadata.query_start_loc_cpu = torch.tensor(
|
|
[0, 10, 20, 30, 40, 50, 60, 70, 80, 90])
|
|
common_attn_metadata.slot_mapping = torch.randn(100, 4, 1024)
|
|
common_attn_metadata.seq_lens_cpu = torch.tensor([2] * 10)
|
|
common_attn_metadata.positions = torch.randn(100)
|
|
common_attn_metadata.attn_mask = None
|
|
common_attn_metadata.attn_state = AscendAttentionState.ChunkedPrefill
|
|
common_attn_metadata.block_table_tensor = torch.randn(100, 4)
|
|
common_attn_metadata.cos = None
|
|
common_attn_metadata.sin = None
|
|
common_attn_metadata.num_input_tokens = 100
|
|
|
|
mock_get_cos_and_sin_mla.return_value = (torch.randn(100),
|
|
torch.randn(100))
|
|
|
|
attn_metadata = builder.build_for_graph_capture(
|
|
common_attn_metadata=common_attn_metadata,
|
|
attn_state=AscendAttentionState.DecodeOnly,
|
|
)
|
|
|
|
assert isinstance(attn_metadata, AscendSFAMetadata)
|
|
assert attn_metadata.attn_state == AscendAttentionState.DecodeOnly
|