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

Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>

[V0 Deprecation] Remove V0 Spec Decode workers (#21152)
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
dd572c0a · Woosuk Kwon · GitHub · 9ffe905a · 9ffe905a · 9ffe905a
Unverified Commit dd572c0a authored Jul 18, 2025 by Woosuk Kwon Committed by GitHub Jul 18, 2025
13 changed files
--- a/vllm/spec_decode/metrics.py
+++ b/vllm/spec_decode/metrics.py
--- a/vllm/spec_decode/mlp_speculator_worker.py
+++ b/vllm/spec_decode/mlp_speculator_worker.py
--- a/vllm/spec_decode/mqa_scorer.py
+++ b/vllm/spec_decode/mqa_scorer.py
--- a/vllm/spec_decode/multi_step_worker.py
+++ b/vllm/spec_decode/multi_step_worker.py
--- a/vllm/spec_decode/ngram_worker.py
+++ b/vllm/spec_decode/ngram_worker.py
--- a/vllm/spec_decode/proposer_worker_base.py
+++ b/vllm/spec_decode/proposer_worker_base.py
--- a/vllm/spec_decode/smaller_tp_proposer_worker.py
+++ b/vllm/spec_decode/smaller_tp_proposer_worker.py
--- a/vllm/spec_decode/spec_decode_worker.py
+++ b/vllm/spec_decode/spec_decode_worker.py
--- a/vllm/spec_decode/target_model_runner.py
+++ b/vllm/spec_decode/target_model_runner.py
--- a/vllm/spec_decode/top1_proposer.py
+++ b/vllm/spec_decode/top1_proposer.py
--- a/vllm/spec_decode/util.py
+++ b/vllm/spec_decode/util.py
--- a/vllm/transformers_utils/configs/eagle.py
+++ b/vllm/transformers_utils/configs/eagle.py
--- a/vllm/worker/worker_base.py
+++ b/vllm/worker/worker_base.py
@@ -397,8 +397,6 @@ class LocalOrDistributedWorkerBase(WorkerBase):

        model_input, worker_input, kwargs = inputs
        num_steps = worker_input.num_steps
-        if execute_model_req is not None and execute_model_req.spec_step_idx:
-            kwargs["spec_step_idx"] = execute_model_req.spec_step_idx

        self.execute_worker(worker_input)