[mypy] Enable type checking for test directory (#5017)
This commit is contained in:
@@ -7,8 +7,8 @@ from vllm.config import SpeculativeConfig
|
||||
from vllm.distributed.communication_op import broadcast_tensor_dict
|
||||
from vllm.logger import init_logger
|
||||
from vllm.model_executor.layers.rejection_sampler import RejectionSampler
|
||||
from vllm.sequence import (ExecuteModelRequest, SamplerOutput,
|
||||
SequenceGroupMetadata)
|
||||
from vllm.sequence import (CompletionSequenceGroupOutput, ExecuteModelRequest,
|
||||
SamplerOutput, SequenceGroupMetadata)
|
||||
from vllm.spec_decode.batch_expansion import BatchExpansionTop1Scorer
|
||||
from vllm.spec_decode.interfaces import (SpeculativeProposals,
|
||||
SpeculativeScorer, SpeculativeScores)
|
||||
@@ -516,13 +516,13 @@ class SpecDecodeWorker(LoraNotSupportedWorkerBase):
|
||||
topk_indices_by_step = topk_indices_by_step.tolist()
|
||||
|
||||
# Construct the output on a per-step, per-sequence basis.
|
||||
sampler_output_list = []
|
||||
sampler_output_list: List[SamplerOutput] = []
|
||||
for step_index in range(num_steps):
|
||||
if all(token_id == -1
|
||||
for token_id in accepted_token_ids_by_step[step_index]):
|
||||
break
|
||||
|
||||
step_output_token_ids = []
|
||||
step_output_token_ids: List[CompletionSequenceGroupOutput] = []
|
||||
for sequence_index in range(batch_size):
|
||||
# Each sequence may have a different num_logprobs; retrieve it.
|
||||
num_logprobs = num_logprobs_per_seq[sequence_index]
|
||||
|
||||
Reference in New Issue
Block a user