fix test

tests/spec_decode/test_dynamic_spec_decode.py

@@ -51,11 +51,14 @@ def test_disable_spec_tokens(queue_size: int, batch_size: int, k: int):
     assert seq_group_metadata_list[
         0].num_speculative_tokens == expected_num_spec_tokens
 
+    draft_worker.sampler_output.side_effect = ValueError(exception_secret)
+
     proposer = Top1Proposer(
         worker=draft_worker,
         device='cpu',  # not used
         vocab_size=100,  # not used
-        max_proposal_len=10,
+        # Must be long enough to avoid being skipped due to length.
+        max_proposal_len=1024,
     )
 
     if queue_size < disable_at_queue_size:
@@ -66,9 +69,9 @@ def test_disable_spec_tokens(queue_size: int, batch_size: int, k: int):
                 num_lookahead_slots=k), )
     else:
         # Should not execute the draft model because spec decode is disabled
-        # for all requests.
+        # for all requests. Accordingly, the proposal length should be 0.
         proposals = proposer.get_proposals(
             execute_model_req=ExecuteModelRequest(
                 seq_group_metadata_list=seq_group_metadata_list,
                 num_lookahead_slots=k), )
-        assert proposals.proposal_lens.tolist() == [k] * batch_size
+        assert proposals.proposal_lens.tolist() == [0] * batch_size

vllm/spec_decode/spec_decode_worker.py

@@ -79,11 +79,8 @@ def create_worker(
             proposer_worker,
             scorer_worker,
             disable_at_queue_size=disable_at_queue_size,
-            # TODO(cade) disable strict mode for speedup.
             rejection_sampler=RejectionSampler(
-                disable_bonus_tokens=disable_bonus_tokens,
-                strict_mode=not envs.VLLM_DISABLE_REJECT_SAMPLING_STRICT_MODE),
-        )
+                disable_bonus_tokens=disable_bonus_tokens, ))
 
     def __init__(
         self,