This reverts commit8966a99710. It breaks the test `tests/e2e/singlecard/spec_decode/test_mtp_eagle_correctness.py::test_deepseek_mtp_correctness[True-FULL_DECODE_ONLY-2-wemaster/deepseek_mtp_main_random_bf16]` - vLLM version: v0.14.0 - vLLM main:d68209402d
This commit is contained in:
@@ -24,15 +24,12 @@ from vllm.forward_context import BatchDescriptor, ForwardContext
|
||||
from tests.ut.base import TestBase
|
||||
from vllm_ascend.attention.attention_v1 import (AscendMetadata,
|
||||
AscendMetadataForDecode)
|
||||
from vllm_ascend.attention.context_parallel.attention_cp import \
|
||||
AscendAttentionCPImpl
|
||||
from vllm_ascend.attention.context_parallel.mla_cp import AscendMlaCPImpl
|
||||
from vllm_ascend.attention.mla_v1 import (AscendMLADecodeMetadata,
|
||||
AscendMLAMetadata)
|
||||
from vllm_ascend.compilation.acl_graph import (
|
||||
ACLGraphEntry, ACLGraphWrapper, get_draft_graph_params, get_graph_params,
|
||||
set_draft_graph_params, set_graph_params,
|
||||
update_draft_graph_params_workspaces)
|
||||
set_draft_graph_params, set_graph_params, update_attn_dcp_pcp_params,
|
||||
update_draft_graph_params_workspaces, update_mla_attn_dcp_pcp_params)
|
||||
|
||||
|
||||
class TestACLGraphEntry(TestBase):
|
||||
@@ -814,9 +811,8 @@ class TestPCPDCPGraphParams(TestBase):
|
||||
out, lse))
|
||||
|
||||
with patch("torch_npu._C._npu_setStream", return_value=None):
|
||||
AscendMlaCPImpl.update_graph_params(
|
||||
self.update_stream, forward_context, 4
|
||||
)
|
||||
update_mla_attn_dcp_pcp_params(self.update_stream, forward_context,
|
||||
4)
|
||||
|
||||
_mock_graph_task_end.assert_called_once()
|
||||
|
||||
@@ -856,8 +852,6 @@ class TestPCPDCPGraphParams(TestBase):
|
||||
out, lse, 2, 0, 0))
|
||||
|
||||
with patch("torch_npu._C._npu_setStream", return_value=None):
|
||||
AscendAttentionCPImpl.update_graph_params(
|
||||
self.update_stream, forward_context, 4, None
|
||||
)
|
||||
update_attn_dcp_pcp_params(self.update_stream, forward_context, 4)
|
||||
|
||||
_mock_graph_task_end.assert_called_once()
|
||||
|
||||
Reference in New Issue
Block a user