[Spec Decode] Rename AR speculators to StandaloneAR / TargetDependentAR (#60335)

Signed-off-by: Giancarlo Delfin <gdelfin@inferact.ai>
This commit is contained in:
Giancarlo Delfin
2026-10-07 14:01:59 -04:00
committed by GitHub
parent fba462e008
commit 6ee9be6b3c
14 changed files with 42 additions and 36 deletions
+2 -2
View File
@@ -23,10 +23,10 @@ from vllm.v1.worker.gpu.cudagraph_utils import BatchExecutionDescriptor
from vllm.v1.worker.gpu.dp_utils import dispatch_cg_and_sync_dp
from vllm.v1.worker.gpu.input_batch import InputBuffers
from vllm.v1.worker.gpu.pcp_manager import PCPManager
from vllm.v1.worker.gpu.spec_decode.autoregressive import (
from vllm.v1.worker.gpu.spec_decode.target_dependent_ar import (
cudagraph_utils as spec_cudagraph_utils,
)
from vllm.v1.worker.gpu.spec_decode.autoregressive.cudagraph_utils import (
from vllm.v1.worker.gpu.spec_decode.target_dependent_ar.cudagraph_utils import (
SpeculatorCudaGraphManager,
)
@@ -22,17 +22,17 @@ from vllm.v1.attention.backends.flash_attn import FlashAttentionMetadata
from vllm.v1.attention.backends.utils import split_decodes_and_prefills
from vllm.v1.worker.gpu.cudagraph_utils import BatchExecutionDescriptor
from vllm.v1.worker.gpu.spec_decode import speculator as base_spec_module
from vllm.v1.worker.gpu.spec_decode.autoregressive import speculator as spec_module
from vllm.v1.worker.gpu.spec_decode.autoregressive.speculator import (
AutoRegressiveSpeculator,
)
from vllm.v1.worker.gpu.spec_decode.multi_module_mtp.speculator import (
MultiModuleMTPSpeculator,
)
from vllm.v1.worker.gpu.spec_decode.speculator import DraftModelSpeculator
from vllm.v1.worker.gpu.spec_decode.target_dependent_ar import speculator as spec_module
from vllm.v1.worker.gpu.spec_decode.target_dependent_ar.speculator import (
TargetDependentARSpeculator,
)
class _TestSpeculator(AutoRegressiveSpeculator):
class _TestSpeculator(TargetDependentARSpeculator):
def load_draft_model(self, target_model, target_attn_layer_names):
return self.test_draft_model