[V0 Deprecation] Remove V0 Spec Decode workers (#21152)

Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
This commit is contained in:
Woosuk Kwon
2025-07-18 21:47:50 -07:00
committed by GitHub
parent 9ffe905a41
commit dd572c0ab3
73 changed files with 191 additions and 14275 deletions
+3 -9
View File
@@ -21,7 +21,6 @@ from vllm.sampling_params import SamplingType
from vllm.sequence import (VLLM_INVALID_TOKEN_ID,
CompletionSequenceGroupOutput, Logprob,
PromptLogprobs, SampleLogprobs, SequenceOutput)
from vllm.spec_decode.metrics import SpecDecodeWorkerMetrics
if envs.VLLM_USE_FLASHINFER_SAMPLER and find_spec("flashinfer"):
# yapf: disable
@@ -119,9 +118,6 @@ class SamplerOutput(
# specified in lieu of prompt token ids or text.
sampled_token_embeds: Optional[torch.Tensor] = None
# Spec decode metrics populated by workers.
spec_decode_worker_metrics: Optional[SpecDecodeWorkerMetrics] = None
# Optional last hidden states from the model.
hidden_states: Optional[torch.Tensor] = None
@@ -159,11 +155,9 @@ class SamplerOutput(
else self.sampled_token_probs.shape)
sampled_token_ids_repr = ("None" if self.sampled_token_ids is None else
self.sampled_token_ids.shape)
return (
f"SamplerOutput(outputs={self.outputs}, "
f"sampled_token_probs={sampled_token_probs_repr}, "
f"sampled_token_ids={sampled_token_ids_repr}, "
f"spec_decode_worker_metrics={self.spec_decode_worker_metrics})")
return (f"SamplerOutput(outputs={self.outputs}, "
f"sampled_token_probs={sampled_token_probs_repr}, "
f"sampled_token_ids={sampled_token_ids_repr})")
class Sampler(nn.Module):