[SpecDecode] Add spec decode support (#500)

### What this PR does / why we need it?
Backport: https://github.com/vllm-project/vllm-ascend/pull/252
This support speculative decoding in Ascend, including speculating with
a draft model、by matching n-grams in the prompt、using MLP speculators
and using EAGLE based draft models.

Backport: https://github.com/vllm-project/vllm-ascend/pull/423
spec decode MultiStepWorker support TP1DraftModelRunner fully, support
run the draft_model_runner with multi-step prepare on the NPU directly
and support draft_model_runner use MLA.

1. before this pr, `MultiStepWorker` would not step into the branch
using NPU prepare, but only into the branch using CPU prepare (`line 52`
of `vllm_ascend/patch/patch_multi_step_worker.py`). Although this has
`no effect` on the `correct operation` of speculative decoding and the
performance of the two branches is basically the same as of the current
version, I support entering this branch in this PR. In general, there
are two main changes in `patch_multi_step_worker.py`: first, the
`is_cuda_like()` check is removed and the `TP1DraftModelRunner`
rewritten in vllm_ascend is used; second, the
`supports_gpu_multi_step()` function is made to return true on NPU
devices when outer Multi_step_worker could work correct.

3. before this pr, `TP1DraftModelRunner` only supports Attention on NPU,
but not MLA. The relevant adaptation is in
`vllm_ascend/worker/draft_model_runner.py`. Although I don’t know why
the `input_positions` of `model_input.attn_metadata` in vllm-ascend
needs to be added in `execute_model`, it is done in `model_runner.py`,
so I also made corresponding changes. Otherwise, when atten_backend is
MLA, it will prompt that input_positions cannot be found.

4. I commented out two lines in `draft_model_runner.py` in `line118` to
support the scenario of K>1.
  ```
  # lora_mapping=model_input.lora_mapping,
  # lora_requests=model_input.lora_requests,
  ```
I added comments. In the future, when vllm-ascend supports lora feature,
the changes here can be restored.

TODO：
- [ ] revert the patch when the related issues are addressed in vllm

### How was this patch tested?
CI passed with new added test.
- e2e test for medusa proposer:
tests/singlecard/spec_decode/e2e/test_medusa_correctness.py
- e2e test for mlp proposer:
tests/singlecard/spec_decode/e2e/test_mlp_correctness.py
- e2e test for n-gram proposer:
tests/singlecard/spec_decode/e2e/test_ngram_correctness.py

Tests for patched files:
- tests/singlecard/spec_decode/test_dynamic_spec_decode.py
- tests/singlecard/spec_decode/test_multi_step_worker.py
- tests/singlecard/spec_decode/test_ngram_worker.py
- tests/singlecard/spec_decode/test_spec_decode_worker.py

---------

Signed-off-by: MengqingCao <cmq0113@163.com>
Co-authored-by: mengwei805 <mengwei25@huawei.com>

tests/singlecard/spec_decode/test_dynamic_spec_decode.py

@@ -0,0 +1,106 @@
+#
+# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
+# This file is a part of the vllm-ascend project.
+# Adapted from vllm-project/vllm/tests/spec_decode/test_dynamic_spec_decode.py
+# Copyright 2023 The vLLM team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+from unittest.mock import MagicMock, patch
+
+import pytest
+import torch
+from vllm.sequence import ExecuteModelRequest
+from vllm.spec_decode.metrics import AsyncMetricsCollector
+from vllm.spec_decode.multi_step_worker import MultiStepWorker
+from vllm.spec_decode.spec_decode_worker import SpecDecodeWorker
+from vllm.spec_decode.top1_proposer import Top1Proposer
+
+from tests.singlecard.spec_decode.test_utils import mock_spec_decode_sampler
+from tests.singlecard.spec_decode.utils import create_batch, mock_worker
+from vllm_ascend.patch.worker import patch_common  # noqa: F401
+
+
+@pytest.mark.parametrize('queue_size', [4])
+@pytest.mark.parametrize('batch_size', [1])
+@pytest.mark.parametrize('k', [1])
+@pytest.mark.parametrize("acceptance_sampler_method",
+                         ["rejection_sampler", "typical_acceptance_sampler"])
+@torch.inference_mode()
+def test_disable_spec_tokens(queue_size: int, batch_size: int, k: int,
+                             acceptance_sampler_method: str):
+    """Verify that speculative tokens are disabled when the batch size
+    exceeds the threshold.
+    """
+    disable_by_batch_size = 3
+    draft_worker = mock_worker(cls=MultiStepWorker)
+    target_worker = mock_worker()
+    metrics_collector = MagicMock(spec=AsyncMetricsCollector)
+    worker = SpecDecodeWorker(proposer_worker=draft_worker,
+                              scorer_worker=target_worker,
+                              spec_decode_sampler=mock_spec_decode_sampler(
+                                  acceptance_sampler_method),
+                              disable_logprobs=False,
+                              metrics_collector=metrics_collector,
+                              disable_by_batch_size=disable_by_batch_size)
+
+    exception_secret = 'artificial stop'
+    draft_worker.get_spec_proposals.side_effect = ValueError(exception_secret)
+
+    seq_group_metadata_list, _, _ = create_batch(batch_size, k)
+    execute_model_req = ExecuteModelRequest(
+        seq_group_metadata_list=seq_group_metadata_list,
+        num_lookahead_slots=k,
+        running_queue_size=queue_size)
+
+    if queue_size > disable_by_batch_size:
+        with patch.object(worker,
+                          '_run_no_spec',
+                          side_effect=ValueError(exception_secret)), \
+            pytest.raises(ValueError, match=exception_secret):
+            worker.execute_model(execute_model_req=execute_model_req)
+
+    # When the batch size is larger than the threshold,
+    # we expect no speculative tokens (0).
+    expected_num_spec_tokens = None if queue_size < disable_by_batch_size else 0
+    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
+        # Must be long enough to avoid being skipped due to length.
+        max_proposal_len=1024,
+    )
+
+    if queue_size < disable_by_batch_size:
+        # Should raise exception when executing the mocked draft model.
+        with pytest.raises(ValueError, match=exception_secret):
+            proposer.get_spec_proposals(
+                execute_model_req=ExecuteModelRequest(
+                    seq_group_metadata_list=seq_group_metadata_list,
+                    num_lookahead_slots=k),
+                seq_ids_with_bonus_token_in_last_step=set())
+    else:
+        # Should not execute the draft model because spec decode is disabled
+        # for all requests. Accordingly, the proposal length should be 0.
+        proposals = proposer.get_spec_proposals(
+            execute_model_req=ExecuteModelRequest(
+                seq_group_metadata_list=seq_group_metadata_list,
+                num_lookahead_slots=k),
+            seq_ids_with_bonus_token_in_last_step=set())
+        assert proposals.proposal_lens.tolist() == [0] * batch_size