[Speculative decoding 3/9] Worker which speculates, scores, and applies rejection sampling (#3103)

.buildkite/test-pipeline.yaml

@@ -28,7 +28,7 @@ steps:
   num_gpus: 2 # only support 1 or 2 for now.
 
 - label: Engine Test
-  command: pytest -v -s engine
+  command: pytest -v -s engine test_sequence.py
 
 - label: Entrypoints Test
   command: pytest -v -s entrypoints
@@ -52,6 +52,9 @@ steps:
 - label: Worker Test
   command: pytest -v -s worker
 
+- label: Speculative decoding tests
+  command: pytest -v -s spec_decode
+
 - label: LoRA Test
   command: pytest -v -s lora --forked
 

tests/spec_decode/test_batch_expansion.py

@@ -0,0 +1,95 @@
+import torch
+import pytest
+
+from vllm.spec_decode.batch_expansion import BatchExpansionTop1Scorer
+
+from .utils import mock_worker, create_seq_group_metadata_from_prompts
+
+
+@pytest.mark.parametrize('num_target_seq_ids', [100])
+def test_create_target_seq_id_iterator(num_target_seq_ids: int):
+    """Verify all new sequence ids are greater than all input
+    seq ids.
+    """
+    scorer = BatchExpansionTop1Scorer(mock_worker(), 'cuda:0', 32_000)
+
+    all_seq_ids = [
+        [1, 3, 5, 7],
+        list(range(100)) + [0],
+        [100],
+    ]
+
+    for seq_ids in all_seq_ids:
+        max_seq_id = max(seq_ids)
+        iterator = scorer._create_target_seq_id_iterator(seq_ids)  # pylint: disable=protected-access
+        for _ in range(num_target_seq_ids):
+            assert next(iterator) > max_seq_id
+
+
+@pytest.mark.parametrize('k', [1, 2, 6])
+def test_get_token_ids_to_score(k: int):
+    """Verify correct tokens are selected for scoring.
+    """
+    proposal_token_ids = torch.tensor(
+        list(range(k)),
+        dtype=torch.int64,
+        device='cuda',
+    )
+
+    expected_output = [
+        [],
+    ]
+    for i in range(proposal_token_ids.shape[0]):
+        expected_output.append(proposal_token_ids[:i + 1].tolist())
+
+    scorer = BatchExpansionTop1Scorer(mock_worker(), 'cuda:0', 32_000)
+    actual_output = scorer._get_token_ids_to_score(proposal_token_ids)  # pylint: disable=protected-access
+
+    actual_output = [
+        x.tolist() if isinstance(x, torch.Tensor) else x for x in actual_output
+    ]
+
+    assert actual_output == expected_output
+
+
+@pytest.mark.parametrize('k', [1, 2, 6])
+def test_create_single_target_seq_group_metadata(k: int):
+    """Verify correct creation of a batch-expanded seq group metadata.
+    """
+
+    prompt_tokens = [1, 2, 3]
+    prev_output_tokens = [4, 5, 6]
+
+    token_ids = list(range(k))
+
+    num_tokens_processed = len(prompt_tokens) + len(prev_output_tokens) - 1
+
+    final_seq_len = len(prompt_tokens) + len(prev_output_tokens) + len(
+        token_ids)
+
+    block_size = 32
+    input_seq_group_metadata = create_seq_group_metadata_from_prompts(
+        [prompt_tokens], 2048 // block_size, block_size, [final_seq_len],
+        [prev_output_tokens], [num_tokens_processed])[0]
+
+    input_seq_id = list(input_seq_group_metadata.seq_data.keys())[0]
+    target_seq_id = 100
+
+    scorer = BatchExpansionTop1Scorer(mock_worker(), 'cuda:0', 32_000)
+    output = scorer._create_single_target_seq_group_metadata(  # pylint: disable=protected-access
+        input_seq_group_metadata,
+        input_seq_id,
+        target_seq_id,
+        token_ids,
+    )
+
+    assert output.request_id == input_seq_group_metadata.request_id
+    assert len(output.seq_data) == 1
+    assert output.seq_data[target_seq_id].get_prompt_token_ids(
+    ) == prompt_tokens
+    assert output.seq_data[target_seq_id].get_output_token_ids(
+    ) == prev_output_tokens + token_ids
+
+    assert len(output.block_tables) == 1
+    assert output.block_tables[
+        target_seq_id] == input_seq_group_metadata.block_tables[input_seq_id]

tests/spec_decode/test_metrics.py

@@ -0,0 +1,157 @@
+import torch
+import math
+import pytest
+
+from unittest.mock import MagicMock
+
+from vllm.spec_decode.metrics import AsyncMetricsCollector
+
+
+def test_initial_call_returns_none():
+    """Expect first call to get metrics to return None.
+    """
+    rej_sampler = MagicMock()
+    rej_sampler.num_accepted_tokens = torch.tensor(0,
+                                                   dtype=torch.long,
+                                                   device='cuda')
+    rej_sampler.num_emitted_tokens = torch.tensor(0,
+                                                  dtype=torch.long,
+                                                  device='cuda')
+    rej_sampler.num_draft_tokens = 0
+
+    collector = AsyncMetricsCollector(rej_sampler)
+    collector.init_gpu_tensors(rank=0)
+    maybe_metrics = collector.maybe_collect_rejsample_metrics(k=5)
+    assert maybe_metrics is None
+
+
+def test_second_call_returns_metrics():
+    """Expect second call to not return None.
+    """
+    rej_sampler = MagicMock()
+    rej_sampler.num_accepted_tokens = torch.tensor(0,
+                                                   dtype=torch.long,
+                                                   device='cuda')
+    rej_sampler.num_emitted_tokens = torch.tensor(0,
+                                                  dtype=torch.long,
+                                                  device='cuda')
+    rej_sampler.num_draft_tokens = 0
+
+    collect_interval_s = 5.0
+    timer = MagicMock()
+    timer.side_effect = [
+        0.0, collect_interval_s + 0.1, collect_interval_s + 0.2
+    ]
+
+    collector = AsyncMetricsCollector(rejection_sampler=rej_sampler,
+                                      timer=timer,
+                                      collect_interval_s=collect_interval_s)
+    collector.init_gpu_tensors(rank=0)
+    _ = collector.maybe_collect_rejsample_metrics(k=5)
+    metrics = collector.maybe_collect_rejsample_metrics(k=5)
+    assert metrics is not None
+
+
+@pytest.mark.parametrize("rank", [1, 2, 3, 4])
+def test_nonzero_rank_noop(rank):
+    """Verify nonzero ranks don't collect metrics.
+    """
+    rej_sampler = MagicMock()
+    rej_sampler.num_accepted_tokens = torch.tensor(0,
+                                                   dtype=torch.long,
+                                                   device='cuda')
+    rej_sampler.num_emitted_tokens = torch.tensor(0,
+                                                  dtype=torch.long,
+                                                  device='cuda')
+    rej_sampler.num_draft_tokens = 0
+
+    collector = AsyncMetricsCollector(rej_sampler)
+    collector.init_gpu_tensors(rank=rank)
+    _ = collector.maybe_collect_rejsample_metrics(k=5)
+    metrics = collector.maybe_collect_rejsample_metrics(k=5)
+    assert metrics is None
+
+
+def test_noop_until_time():
+    """Verify metrics aren't collected until enough time passes.
+    """
+    rej_sampler = MagicMock()
+    rej_sampler.num_accepted_tokens = torch.tensor(0,
+                                                   dtype=torch.long,
+                                                   device='cuda')
+    rej_sampler.num_emitted_tokens = torch.tensor(0,
+                                                  dtype=torch.long,
+                                                  device='cuda')
+    rej_sampler.num_draft_tokens = 0
+
+    collect_interval_s = 5.0
+    timer = MagicMock()
+    timer.side_effect = [
+        0.0, collect_interval_s - 0.1, collect_interval_s - 0.1,
+        collect_interval_s + 0.1, collect_interval_s + 0.1
+    ]
+
+    collector = AsyncMetricsCollector(rejection_sampler=rej_sampler,
+                                      timer=timer,
+                                      collect_interval_s=collect_interval_s)
+    collector.init_gpu_tensors(rank=0)
+
+    _ = collector.maybe_collect_rejsample_metrics(k=5)
+    metrics = collector.maybe_collect_rejsample_metrics(k=5)
+    assert metrics is None
+
+    _ = collector.maybe_collect_rejsample_metrics(k=5)
+    metrics = collector.maybe_collect_rejsample_metrics(k=5)
+    assert metrics is not None
+
+
+@pytest.mark.parametrize("has_data", [True, False])
+def test_initial_metrics_has_correct_values(has_data: bool):
+    """Test correctness of metrics data.
+    """
+    if has_data:
+        num_accepted_tokens = 103
+        num_emitted_tokens = 104
+        num_draft_tokens = 105
+    else:
+        num_accepted_tokens = 0
+        num_emitted_tokens = 0
+        num_draft_tokens = 0
+    k = 5
+
+    num_possible_tokens = AsyncMetricsCollector.get_max_num_accepted_tokens(
+        num_draft_tokens, k)
+
+    rej_sampler = MagicMock()
+    rej_sampler.num_accepted_tokens = torch.tensor(num_accepted_tokens,
+                                                   dtype=torch.long,
+                                                   device='cuda')
+    rej_sampler.num_emitted_tokens = torch.tensor(num_emitted_tokens,
+                                                  dtype=torch.long,
+                                                  device='cuda')
+    rej_sampler.num_draft_tokens = num_draft_tokens
+
+    collect_interval_s = 5.0
+    timer = MagicMock()
+    timer.side_effect = [
+        0.0, collect_interval_s + 0.1, collect_interval_s + 0.2
+    ]
+
+    collector = AsyncMetricsCollector(rejection_sampler=rej_sampler,
+                                      timer=timer,
+                                      collect_interval_s=collect_interval_s)
+    collector.init_gpu_tensors(rank=0)
+    _ = collector.maybe_collect_rejsample_metrics(k)
+    metrics = collector.maybe_collect_rejsample_metrics(k)
+
+    assert metrics.num_spec_tokens == k
+    assert metrics.accepted_tokens == num_accepted_tokens
+    assert metrics.draft_tokens == num_draft_tokens
+    assert metrics.emitted_tokens == num_emitted_tokens
+
+    if has_data:
+        assert metrics.draft_acceptance_rate == num_accepted_tokens / num_draft_tokens
+        assert metrics.system_efficiency == num_emitted_tokens / num_possible_tokens
+    else:
+        assert math.isnan(metrics.draft_acceptance_rate)
+        assert math.isnan(metrics.system_efficiency)

tests/spec_decode/test_multi_step_worker.py

@@ -3,14 +3,15 @@ import random
 import pytest
 from unittest.mock import MagicMock
 
-from vllm.worker.spec_decode.multi_step_worker import MultiStepWorker
+from vllm.spec_decode.multi_step_worker import MultiStepWorker, DraftModelTop1Proposer
 from vllm.worker.worker import Worker
 from vllm.model_executor.utils import set_random_seed
+from vllm.sequence import SamplerOutput
 
 from .utils import (create_execute_model_data, create_worker,
                     create_seq_group_metadata_from_prompts, zero_kv_cache,
                     patch_execute_model_with_seeds,
-                    assert_logprobs_dict_allclose)
+                    assert_logprobs_dict_allclose, create_batch)
 
 
 @pytest.mark.parametrize('num_steps', list(range(1, 17)))
@@ -259,3 +260,160 @@ def test_same_output_for_multi_step():
             multi_step_output_logprobs, single_step_output_logprobs):
         assert_logprobs_dict_allclose(multi_step_logprobs,
                                       single_step_logprobs)
+
+
+@torch.inference_mode()
+def test_draft_proposals_full_speculation_len():
+    """Verify DraftModelTop1Proposer correctly handles case where all sequences
+    can speculate.
+    """
+    k = 10
+    batch_size = 32
+    vocab_size = 32_000
+    device = 'cuda:0'
+
+    draft_worker = MagicMock()
+    proposer = DraftModelTop1Proposer(
+        draft_worker=draft_worker,
+        device=device,
+        max_model_len=2048,
+        vocab_size=vocab_size,
+    )
+    draft_worker.execute_model_multi_step.return_value = [
+        SamplerOutput(
+            outputs=[],
+            sampled_token_probs=torch.rand(batch_size,
+                                           vocab_size,
+                                           device=device,
+                                           dtype=torch.float32),
+            sampled_token_ids=torch.randint(low=0,
+                                            high=vocab_size,
+                                            size=(batch_size, ),
+                                            device=device,
+                                            dtype=torch.long),
+        ) for _ in range(k)
+    ]
+
+    execute_model_data, _, _ = create_batch(batch_size, k)
+
+    proposals = proposer.get_proposals(
+        **execute_model_data.to_dict(),
+        max_proposal_len=k,
+    )
+
+    assert torch.is_tensor(proposals.proposal_token_ids)
+    assert torch.is_tensor(proposals.proposal_probs)
+
+    assert proposals.proposal_token_ids.shape == torch.Size([batch_size, k])
+    assert proposals.proposal_probs.shape[:-1] == torch.Size([batch_size, k])
+
+    assert proposals.proposal_lens.shape == torch.Size([batch_size])
+    assert proposals.proposal_lens.tolist() == [k for _ in range(batch_size)]
+
+
+@torch.inference_mode()
+def test_draft_proposals_no_speculations():
+    """Verify DraftModelTop1Proposer correctly handles case where no sequences
+    can speculate.
+    """
+    k = 10
+    batch_size = 32
+    vocab_size = 32_000
+    device = 'cuda:0'
+    prompt_len = 10
+
+    draft_worker = MagicMock()
+    proposer = DraftModelTop1Proposer(
+        draft_worker=draft_worker,
+        device=device,
+        max_model_len=prompt_len + k - 1,
+        vocab_size=vocab_size,
+    )
+
+    execute_model_data, _, _ = create_batch(batch_size,
+                                            k,
+                                            prompt_len=prompt_len)
+
+    proposals = proposer.get_proposals(
+        **execute_model_data.to_dict(),
+        max_proposal_len=k,
+    )
+
+    assert torch.is_tensor(proposals.proposal_token_ids)
+    assert torch.is_tensor(proposals.proposal_probs)
+
+    assert proposals.proposal_token_ids.shape == torch.Size([0, k])
+    assert proposals.proposal_probs.shape[:-1] == torch.Size([0, k])
+
+    assert proposals.proposal_lens.shape == torch.Size([batch_size])
+    assert proposals.proposal_lens.tolist() == [0 for _ in range(batch_size)]
+
+
+@torch.inference_mode()
+def test_draft_proposals_mixed_k():
+    """Verify DraftModelTop1Proposer correctly handles case some sequences can
+    speculate and some can't.
+    """
+    k = 10
+    batch_size = 32
+    vocab_size = 32_000
+    device = 'cuda:0'
+
+    small_prompt_len = 5
+    long_prompt_len = 10
+    prev_output_token_len = 20
+
+    expected_num_proposal_seqs = 6
+    expected_num_no_proposal_seqs = batch_size - expected_num_proposal_seqs
+
+    prompt_len = [
+        small_prompt_len for _ in range(expected_num_proposal_seqs - 1)
+    ] + [long_prompt_len
+         for _ in range(expected_num_no_proposal_seqs)] + [small_prompt_len]
+
+    draft_worker = MagicMock()
+    proposer = DraftModelTop1Proposer(
+        draft_worker=draft_worker,
+        device=device,
+        max_model_len=long_prompt_len + prev_output_token_len + k - 1,
+        vocab_size=vocab_size,
+    )
+
+    draft_worker.execute_model_multi_step.return_value = [
+        SamplerOutput(
+            outputs=[],
+            sampled_token_probs=torch.rand(expected_num_proposal_seqs,
+                                           vocab_size,
+                                           device=device,
+                                           dtype=torch.float32),
+            sampled_token_ids=torch.randint(
+                low=0,
+                high=vocab_size,
+                size=(expected_num_proposal_seqs, ),
+                device=device,
+                dtype=torch.long),
+        ) for _ in range(k)
+    ]
+
+    execute_model_data, _, _ = create_batch(
+        batch_size,
+        k,
+        prompt_len=prompt_len,
+        prev_output_token_len=prev_output_token_len,
+    )
+
+    proposals = proposer.get_proposals(
+        **execute_model_data.to_dict(),
+        max_proposal_len=k,
+    )
+
+    assert torch.is_tensor(proposals.proposal_token_ids)
+    assert torch.is_tensor(proposals.proposal_probs)
+
+    assert proposals.proposal_token_ids.shape == torch.Size([batch_size, k])
+    assert proposals.proposal_probs.shape[:-1] == torch.Size([batch_size, k])
+
+    assert proposals.proposal_lens.shape == torch.Size([batch_size])
+    assert proposals.proposal_lens.tolist() == [
+        k for _ in range(expected_num_proposal_seqs - 1)
+    ] + [0 for _ in range(expected_num_no_proposal_seqs)] + [k]

tests/spec_decode/test_spec_decode_worker.py

@@ -0,0 +1,591 @@
+import torch
+import random
+import pytest
+from unittest.mock import MagicMock
+
+from vllm.spec_decode.multi_step_worker import MultiStepWorker
+from vllm.spec_decode.spec_decode_worker import SpecDecodeWorker, split_num_cache_blocks_evenly
+from vllm.spec_decode.interfaces import SpeculativeProposals
+from vllm.model_executor.utils import set_random_seed
+from vllm.model_executor.layers.rejection_sampler import RejectionSampler
+from .utils import mock_worker, create_batch, ExecuteModelData, create_sampler_output_list
+from vllm.spec_decode.metrics import SpecDecodeWorkerMetrics, AsyncMetricsCollector
+
+
+@pytest.mark.parametrize('k', [1, 2, 6])
+@pytest.mark.parametrize('batch_size', [1, 2, 32])
+@torch.inference_mode()
+def test_correctly_calls_draft_model(k: int, batch_size: int):
+    """Verify SpecDecodeWorker calls the draft worker with correct
+    inputs. Everything else is mocked out.
+    """
+    draft_worker = mock_worker(cls=MultiStepWorker)
+    target_worker = mock_worker()
+    rejection_sampler = MagicMock(spec=RejectionSampler)
+    metrics_collector = MagicMock(spec=AsyncMetricsCollector)
+    worker = SpecDecodeWorker(draft_worker, target_worker, rejection_sampler,
+                              metrics_collector)
+
+    exception_secret = 'artifical stop'
+    draft_worker.get_spec_proposals.side_effect = ValueError(exception_secret)
+
+    execute_model_data, _, _ = create_batch(batch_size, k)
+
+    with pytest.raises(ValueError, match=exception_secret):
+        worker.execute_model(**execute_model_data.to_dict(), num_spec_tokens=k)
+
+    call_args_list = draft_worker.get_spec_proposals.call_args_list
+    assert len(call_args_list) == 1
+
+    for args, _ in call_args_list:
+        (seq_group_metadata_list, blocks_to_swap_in, blocks_to_swap_out,
+         blocks_to_copy, actual_k) = args
+        actual_execute_model_data = ExecuteModelData(seq_group_metadata_list,
+                                                     blocks_to_swap_in,
+                                                     blocks_to_swap_out,
+                                                     blocks_to_copy)
+        assert actual_execute_model_data == execute_model_data
+        assert actual_k == k
+
+
+@pytest.mark.parametrize('k', [1, 2, 6])
+@pytest.mark.parametrize('batch_size', [1, 2, 32])
+@torch.inference_mode()
+def test_correctly_calls_target_model(k: int, batch_size: int):
+    """Verify SpecDecodeWorker calls the target model with correct
+    inputs. Everything else is mocked out.
+    """
+    draft_worker = mock_worker(cls=MultiStepWorker)
+    target_worker = mock_worker()
+    rejection_sampler = MagicMock(spec=RejectionSampler)
+    rejection_sampler.token_id_dtype = torch.int64
+    metrics_collector = MagicMock(spec=AsyncMetricsCollector)
+
+    draft_worker.device = 'cuda'
+    target_worker.device = 'cuda'
+
+    set_random_seed(1)
+
+    worker = SpecDecodeWorker(draft_worker, target_worker, rejection_sampler,
+                              metrics_collector)
+    worker.init_model()
+
+    vocab_size = 32_000
+
+    proposal_token_ids = torch.randint(low=0,
+                                       high=vocab_size,
+                                       size=(batch_size, k),
+                                       dtype=torch.int64,
+                                       device='cuda')
+    proposal_probs = torch.rand(batch_size,
+                                k,
+                                vocab_size,
+                                dtype=torch.float32,
+                                device='cuda')
+    proposal_lens = torch.ones(batch_size, dtype=torch.int64,
+                               device='cuda') * k
+
+    execute_model_data, prompts, prev_output_tokens = create_batch(
+        batch_size, k)
+
+    draft_worker.get_spec_proposals.return_value = SpeculativeProposals(
+        proposal_token_ids=proposal_token_ids,
+        proposal_probs=proposal_probs,
+        proposal_lens=proposal_lens)
+
+    exception_secret = 'artifical stop'
+    target_worker.execute_model.side_effect = ValueError(exception_secret)
+
+    with pytest.raises(ValueError, match=exception_secret):
+        worker.execute_model(**execute_model_data.to_dict(), num_spec_tokens=k)
+
+    seen_contexts = []
+
+    call_args_list = target_worker.execute_model.call_args_list
+    assert len(call_args_list) == 1
+    for args, kwargs in call_args_list:
+        target_execute_model_data = ExecuteModelData.from_dict(kwargs)
+
+        assert len(target_execute_model_data.seq_group_metadata_list) == (
+            k + 1) * batch_size
+        for seq_group_metadata in (
+                target_execute_model_data.seq_group_metadata_list):
+            for seq_data in seq_group_metadata.seq_data.values():
+                seen_contexts.append(seq_data.get_token_ids())
+
+    expected_seen_contexts = []
+
+    for prompt, prev_generated, draft_tokens in zip(
+            prompts, prev_output_tokens, proposal_token_ids.tolist()):
+
+        for i in range(len(draft_tokens) + 1):
+            expected_seen_contexts.append(prompt + prev_generated +
+                                          draft_tokens[:i])
+
+    seen_contexts.sort()
+    expected_seen_contexts.sort()
+    assert expected_seen_contexts == seen_contexts
+
+
+@pytest.mark.parametrize('k', [1, 2, 6])
+@pytest.mark.parametrize('batch_size', [1, 2, 32])
+@torch.inference_mode()
+def test_correctly_calls_rejection_sampler(k: int, batch_size: int):
+    """Verify SpecDecodeWorker calls the rejection sampler with
+    correct inputs. Everything else is mocked out.
+    """
+    vocab_size = 32_000
+
+    draft_worker = mock_worker(cls=MultiStepWorker, vocab_size=vocab_size)
+    target_worker = mock_worker(vocab_size=vocab_size)
+    rejection_sampler = MagicMock(spec=RejectionSampler)
+    rejection_sampler.token_id_dtype = torch.int64
+    metrics_collector = MagicMock(spec=AsyncMetricsCollector)
+    draft_worker.device = 'cuda'
+    target_worker.device = 'cuda'
+
+    set_random_seed(1)
+
+    worker = SpecDecodeWorker(draft_worker, target_worker, rejection_sampler,
+                              metrics_collector)
+    worker.init_model()
+
+    proposal_token_ids = torch.randint(low=0,
+                                       high=vocab_size,
+                                       size=(batch_size, k),
+                                       dtype=torch.int64,
+                                       device='cuda')
+    proposal_probs = torch.rand(batch_size,
+                                k,
+                                vocab_size,
+                                dtype=torch.float32,
+                                device='cuda')
+
+    proposal_lens = torch.ones(batch_size, dtype=torch.int64,
+                               device='cuda') * k
+
+    execute_model_data, _, _ = create_batch(batch_size, k)
+
+    draft_worker.get_spec_proposals.return_value = SpeculativeProposals(
+        proposal_token_ids=proposal_token_ids,
+        proposal_probs=proposal_probs,
+        proposal_lens=proposal_lens)
+
+    target_token_ids = torch.randint(low=0,
+                                     high=vocab_size,
+                                     size=(1, batch_size * (k + 1)),
+                                     dtype=torch.int64,
+                                     device='cuda')
+    target_token_probs = torch.rand(1,
+                                    batch_size * (k + 1),
+                                    vocab_size,
+                                    dtype=torch.float32,
+                                    device='cuda')
+    target_output = create_sampler_output_list(target_token_ids,
+                                               target_token_probs)
+
+    target_worker.execute_model.return_value = target_output[0]
+
+    exception_secret = 'artifical stop'
+    rejection_sampler.side_effect = ValueError(exception_secret)
+
+    with pytest.raises(ValueError, match=exception_secret):
+        worker.execute_model(**execute_model_data.to_dict(), num_spec_tokens=k)
+
+    assert len(rejection_sampler.call_args_list) == 1
+    args, _ = rejection_sampler.call_args_list[0]
+    (actual_proposal_scores, actual_bonus_token_ids, actual_proposal_probs,
+     actual_proposal_token_ids) = args
+
+    assert torch.equal(actual_bonus_token_ids,
+                       target_token_ids.reshape(batch_size, k + 1)[:, -1:])
+    assert torch.equal(
+        actual_proposal_scores,
+        target_token_probs.reshape(batch_size, k + 1, -1)[:, :-1])
+    assert torch.equal(actual_proposal_token_ids, proposal_token_ids)
+    assert torch.equal(actual_proposal_probs, proposal_probs)
+
+
+@pytest.mark.parametrize('k', [1, 2, 6])
+@pytest.mark.parametrize('batch_size', [1, 2, 32])
+@torch.inference_mode()
+def test_correctly_formats_output(k: int, batch_size: int):
+    """Verify SpecDecodeWorker formats sampler output correctly.
+    Everything else is mocked out.
+    """
+    vocab_size = 32_000
+
+    draft_worker = mock_worker(cls=MultiStepWorker, vocab_size=vocab_size)
+    target_worker = mock_worker(vocab_size=vocab_size)
+    rejection_sampler = MagicMock(spec=RejectionSampler)
+    rejection_sampler.token_id_dtype = torch.int64
+    metrics_collector = MagicMock(spec=AsyncMetricsCollector)
+    draft_worker.device = 'cuda'
+    target_worker.device = 'cuda'
+
+    set_random_seed(1)
+
+    worker = SpecDecodeWorker(draft_worker, target_worker, rejection_sampler,
+                              metrics_collector)
+    worker.init_model()
+
+    proposal_token_ids = torch.randint(low=0,
+                                       high=vocab_size,
+                                       size=(batch_size, k),
+                                       dtype=torch.int64,
+                                       device='cuda')
+    proposal_probs = torch.rand(batch_size,
+                                k,
+                                vocab_size,
+                                dtype=torch.float32,
+                                device='cuda')
+
+    proposal_lens = torch.ones(batch_size, dtype=torch.int64,
+                               device='cuda') * k
+
+    execute_model_data, _, _ = create_batch(batch_size, k)
+
+    draft_worker.get_spec_proposals.return_value = SpeculativeProposals(
+        proposal_token_ids=proposal_token_ids,
+        proposal_probs=proposal_probs,
+        proposal_lens=proposal_lens)
+
+    target_token_ids = torch.randint(low=0,
+                                     high=vocab_size,
+                                     size=(1, batch_size * (k + 1)),
+                                     dtype=torch.int64,
+                                     device='cuda')
+    target_token_probs = torch.rand(1,
+                                    batch_size * (k + 1),
+                                    vocab_size,
+                                    dtype=torch.float32,
+                                    device='cuda')
+    target_output = create_sampler_output_list(target_token_ids,
+                                               target_token_probs)
+
+    target_worker.execute_model.return_value = target_output[0]
+
+    rejection_sampler_output = torch.randint(low=0,
+                                             high=vocab_size,
+                                             size=(batch_size, k + 1),
+                                             dtype=torch.int64,
+                                             device='cuda')
+    for i in range(batch_size):
+        minimum_accepted_tokens = 1
+        rejection_sampler_output[i][
+            -random.randint(minimum_accepted_tokens, k + 1):] = -1
+
+    rejection_sampler.return_value = rejection_sampler_output
+
+    output = worker.execute_model(**execute_model_data.to_dict(),
+                                  num_spec_tokens=k)
+
+    expected_output = create_sampler_output_list(
+        rejection_sampler_output.transpose(0, 1), [None for _ in range(k + 1)])
+
+    seq_ids = [
+        next(iter(seq_group_metadata.seq_data.keys()))
+        for seq_group_metadata in execute_model_data.seq_group_metadata_list
+    ]
+    actual_output_by_seq = {seq_id: [] for seq_id in seq_ids}
+    expected_output_by_seq = {seq_id: [] for seq_id in seq_ids}
+
+    for step in output:
+        for seq_group in step:
+            for sample in seq_group.samples:
+                seq_id = sample.parent_seq_id
+                actual_output_by_seq[seq_id].append(sample)
+
+    for step in expected_output:
+        for seq_group in step:
+            for sample in seq_group.samples:
+                seq_id = sample.parent_seq_id
+                expected_output_by_seq[seq_id].append(sample)
+
+    all_seen_seq_ids = set(
+        list(actual_output_by_seq.keys()) +
+        list(expected_output_by_seq.keys()))
+    for seq_id in all_seen_seq_ids:
+        actual_by_step = actual_output_by_seq[seq_id]
+        expected_by_step = expected_output_by_seq[seq_id]
+
+        for i in range(k + 1):
+            if i >= len(actual_by_step):
+                assert expected_by_step[i].output_token == -1
+                continue
+            assert actual_by_step[i].output_token == expected_by_step[
+                i].output_token
+            assert actual_by_step[i].logprobs == expected_by_step[i].logprobs
+
+
+@pytest.mark.parametrize('k', [1, 2])
+@pytest.mark.parametrize('batch_size', [1])
+@pytest.mark.parametrize('returns_metrics', [True, False])
+@torch.inference_mode()
+def test_collects_metrics(k: int, batch_size: int, returns_metrics: bool):
+    """Verify SpecDecodeWorker collects metrics.
+    """
+    vocab_size = 32_000
+
+    draft_worker = mock_worker(cls=MultiStepWorker, vocab_size=vocab_size)
+    target_worker = mock_worker(vocab_size=vocab_size)
+    rejection_sampler = MagicMock(spec=RejectionSampler)
+    rejection_sampler.token_id_dtype = torch.int64
+    metrics_collector = MagicMock(spec=AsyncMetricsCollector)
+    draft_worker.device = 'cuda'
+    target_worker.device = 'cuda'
+
+    set_random_seed(1)
+
+    worker = SpecDecodeWorker(draft_worker, target_worker, rejection_sampler,
+                              metrics_collector)
+    worker.init_model()
+
+    proposal_token_ids = torch.randint(low=0,
+                                       high=vocab_size,
+                                       size=(batch_size, k),
+                                       dtype=torch.int64,
+                                       device='cuda')
+    proposal_probs = torch.rand(batch_size,
+                                k,
+                                vocab_size,
+                                dtype=torch.float32,
+                                device='cuda')
+
+    proposal_lens = torch.ones(batch_size, dtype=torch.int64,
+                               device='cuda') * k
+
+    execute_model_data, _, _ = create_batch(batch_size, k)
+
+    draft_worker.get_spec_proposals.return_value = SpeculativeProposals(
+        proposal_token_ids=proposal_token_ids,
+        proposal_probs=proposal_probs,
+        proposal_lens=proposal_lens)
+
+    target_token_ids = torch.randint(low=0,
+                                     high=vocab_size,
+                                     size=(1, batch_size * (k + 1)),
+                                     dtype=torch.int64,
+                                     device='cuda')
+    target_token_probs = torch.rand(1,
+                                    batch_size * (k + 1),
+                                    vocab_size,
+                                    dtype=torch.float32,
+                                    device='cuda')
+    target_output = create_sampler_output_list(target_token_ids,
+                                               target_token_probs)
+
+    target_worker.execute_model.return_value = target_output[0]
+
+    rejection_sampler_output = torch.randint(low=0,
+                                             high=vocab_size,
+                                             size=(batch_size, k + 1),
+                                             dtype=torch.int64,
+                                             device='cuda')
+    for i in range(batch_size):
+        minimum_accepted_tokens = 1
+        rejection_sampler_output[i][
+            -random.randint(minimum_accepted_tokens, k + 1):] = -1
+
+    rejection_sampler.return_value = rejection_sampler_output
+
+    mock_rejsample_metrics = MagicMock(
+        spec=SpecDecodeWorkerMetrics) if returns_metrics else None
+    metrics_collector.maybe_collect_rejsample_metrics.return_value = mock_rejsample_metrics
+
+    output = worker.execute_model(**execute_model_data.to_dict(),
+                                  num_spec_tokens=k)
+    assert output[0].spec_decode_worker_metrics == mock_rejsample_metrics
+
+    call_args_list = metrics_collector.maybe_collect_rejsample_metrics.call_args_list
+    assert len(call_args_list) == 1
+    args, kwargs = call_args_list[0]
+    assert args[0] == k or kwargs.get('k', -1) == k
+
+
+@pytest.mark.parametrize('k', [0])
+@pytest.mark.parametrize('batch_size', [1, 2, 32])
+@torch.inference_mode()
+def test_k_equals_zero(k: int, batch_size: int):
+    """Verify that the SpecDecodeWorker calls the draft and target workers
+    when k is zero. This happens during prefill.
+    """
+    draft_worker = mock_worker(cls=MultiStepWorker)
+    target_worker = mock_worker()
+    rejection_sampler = MagicMock(spec=RejectionSampler)
+    rejection_sampler.token_id_dtype = torch.int64
+    metrics_collector = MagicMock(spec=AsyncMetricsCollector)
+
+    draft_worker.device = 'cuda'
+    target_worker.device = 'cuda'
+
+    set_random_seed(1)
+
+    worker = SpecDecodeWorker(draft_worker, target_worker, rejection_sampler,
+                              metrics_collector)
+
+    execute_model_data, prompts, prev_output_tokens = create_batch(
+        batch_size, k, prev_output_token_len=0)
+
+    out = worker.execute_model(**execute_model_data.to_dict(),
+                               num_spec_tokens=k)
+
+    assert len(out) == 1, f"expected only one token output when {k=}"
+    assert out[0].probs is None, "expect gpu tensor references to be None"
+    assert out[
+        0].sampled_tokens is None, "expect gpu tensor references to be None"
+
+    draft_worker.execute_model.assert_called_once_with(
+        **execute_model_data.to_dict(), return_python_output=False)
+    target_worker.execute_model.assert_called_once_with(
+        **execute_model_data.to_dict())
+
+
+@pytest.mark.parametrize('k', [0, 5])
+@pytest.mark.parametrize('batch_size', [0])
+@torch.inference_mode()
+def test_empty_input_batch(k: int, batch_size: int):
+    """Verify that the SpecDecodeWorker calls the draft and target workers
+    when the input batch is empty. This can happen if the engine communicates
+    to the workers information without scheduling a batch.
+    """
+    draft_worker = mock_worker(cls=MultiStepWorker)
+    target_worker = mock_worker()
+    rejection_sampler = MagicMock(spec=RejectionSampler)
+    rejection_sampler.token_id_dtype = torch.int64
+    metrics_collector = MagicMock(spec=AsyncMetricsCollector)
+
+    draft_worker.device = 'cuda'
+    target_worker.device = 'cuda'
+
+    set_random_seed(1)
+
+    worker = SpecDecodeWorker(draft_worker, target_worker, rejection_sampler,
+                              metrics_collector)
+
+    execute_model_data, prompts, prev_output_tokens = create_batch(
+        batch_size, k, prev_output_token_len=0)
+
+    out = worker.execute_model(**execute_model_data.to_dict(),
+                               num_spec_tokens=k)
+
+    assert len(out) == 1, f"expected only one token output when {k=}"
+    assert out[0].probs is None, "expect gpu tensor references to be None"
+    assert out[
+        0].sampled_tokens is None, "expect gpu tensor references to be None"
+
+    draft_worker.execute_model.assert_called_once_with(
+        **execute_model_data.to_dict(), return_python_output=False)
+    target_worker.execute_model.assert_called_once_with(
+        **execute_model_data.to_dict())
+
+
+@torch.inference_mode()
+def test_init_model():
+    """Verify SpecDecodeWorker invokes proposer/scorer worker init_model, as
+    well as other GPU initialization.
+    """
+    draft_worker = mock_worker(cls=MultiStepWorker)
+    target_worker = mock_worker()
+    rejection_sampler = MagicMock(spec=RejectionSampler)
+    rejection_sampler.token_id_dtype = torch.int64
+    metrics_collector = MagicMock(spec=AsyncMetricsCollector)
+
+    worker = SpecDecodeWorker(draft_worker, target_worker, rejection_sampler,
+                              metrics_collector)
+
+    worker.init_model()
+
+    draft_worker.init_model.assert_called_once()
+
+    target_worker.init_model.assert_called_once()
+
+    metrics_collector.init_gpu_tensors.assert_called_once()
+    rejection_sampler.init_gpu_tensors.assert_called_once()
+
+
+@torch.inference_mode()
+def test_init_cache_engine():
+    """Verify SpecDecodeWorker invokes init_cache_engine on proposer/scorer
+    workers.
+    """
+    draft_worker = mock_worker(cls=MultiStepWorker)
+    target_worker = mock_worker()
+    rejection_sampler = MagicMock(spec=RejectionSampler)
+    rejection_sampler.token_id_dtype = torch.int64
+    metrics_collector = MagicMock(spec=AsyncMetricsCollector)
+
+    worker = SpecDecodeWorker(draft_worker, target_worker, rejection_sampler,
+                              metrics_collector)
+
+    cache_config = MagicMock()
+
+    worker.init_cache_engine(cache_config)
+
+    draft_worker.init_cache_engine.assert_called_once_with(cache_config)
+    target_worker.init_cache_engine.assert_called_once_with(cache_config)
+
+
+@pytest.mark.parametrize('available_gpu_blocks', [1, 1024])
+@pytest.mark.parametrize('available_cpu_blocks', [500])
+@pytest.mark.parametrize('target_cache_block_size_bytes', [2 * 2 * 4096])
+@pytest.mark.parametrize('draft_kv_size_bytes', [0, 2 * 2 * 768, 2 * 2 * 4096])
+@torch.inference_mode()
+def test_profile_num_available_blocks(available_gpu_blocks: int,
+                                      available_cpu_blocks: int,
+                                      target_cache_block_size_bytes: int,
+                                      draft_kv_size_bytes: int):
+    """Verify SpecDecodeWorker correctly profiles num available GPU blocks.
+    Specifically, it should run profiling in the scorer worker, and then evenly
+    split the blocks between proposer and scorer worker.
+    """
+    draft_worker = mock_worker(cls=MultiStepWorker)
+    target_worker = mock_worker()
+    rejection_sampler = MagicMock(spec=RejectionSampler)
+    rejection_sampler.token_id_dtype = torch.int64
+    metrics_collector = MagicMock(spec=AsyncMetricsCollector)
+
+    target_worker.profile_num_available_blocks.return_value = (
+        available_gpu_blocks, available_cpu_blocks)
+    target_worker.get_cache_block_size_bytes.return_value = target_cache_block_size_bytes
+    draft_worker.get_cache_block_size_bytes.return_value = draft_kv_size_bytes
+
+    worker = SpecDecodeWorker(draft_worker, target_worker, rejection_sampler,
+                              metrics_collector)
+
+    # These values do not directly impact the adjusted block size calculation,
+    # so they can be fixed.
+    gpu_memory_utilization = 0.9
+    cpu_swap_space = 100
+    block_size = 16
+
+    num_gpu_blocks, num_cpu_blocks = worker.profile_num_available_blocks(
+        block_size, gpu_memory_utilization, cpu_swap_space, cache_dtype="auto")
+
+    target_worker.profile_num_available_blocks.assert_called_once_with(
+        block_size, gpu_memory_utilization, cpu_swap_space, "auto")
+    assert num_cpu_blocks == available_cpu_blocks
+
+    assert num_gpu_blocks == split_num_cache_blocks_evenly(
+        target_cache_block_size_bytes, draft_kv_size_bytes,
+        available_gpu_blocks)
+
+
+@pytest.mark.parametrize('available_gpu_blocks',
+                         list(range(20)) + [1024, 1024**2])
+@pytest.mark.parametrize('target_cache_block_size_bytes',
+                         [2 * 2 * 4096, 2 * 2 * 8192])
+@pytest.mark.parametrize('draft_kv_size_bytes', [0, 2 * 2 * 768, 2 * 2 * 4096])
+@torch.inference_mode()
+def test_split_num_cache_blocks_evenly(available_gpu_blocks: int,
+                                       target_cache_block_size_bytes: int,
+                                       draft_kv_size_bytes: int):
+    """Verify split_num_cache_blocks_evenly does not exceed original memory
+    allocation in bytes.
+    """
+    num_blocks = split_num_cache_blocks_evenly(target_cache_block_size_bytes,
+                                               draft_kv_size_bytes,
+                                               available_gpu_blocks)
+    assert (num_blocks * target_cache_block_size_bytes) + (
+        num_blocks * draft_kv_size_bytes) <= (available_gpu_blocks *
+                                              target_cache_block_size_bytes)

tests/spec_decode/test_utils.py

@@ -0,0 +1,111 @@
+from vllm.spec_decode.util import get_all_seq_ids
+from vllm.sequence import SequenceGroupMetadata
+from vllm.spec_decode.util import split_batch_by_proposal_len
+
+import pytest
+from unittest.mock import MagicMock
+
+
+def test_get_all_seq_ids():
+    """Verify get_all_seq_ids extracts all seq ids.
+    """
+    expected_seq_ids = list(range(10)) + list(range(100, 110))
+
+    seq_group_metadata_list = [
+        SequenceGroupMetadata(
+            request_id=str(seq_id),
+            is_prompt=True,
+            seq_data={
+                seq_id: MagicMock(),
+            },
+            sampling_params=MagicMock(),
+            block_tables={
+                seq_id: MagicMock(),
+            },
+            lora_request=None,
+        ) for seq_id in expected_seq_ids
+    ]
+
+    actual_seq_ids = get_all_seq_ids(seq_group_metadata_list)
+    assert actual_seq_ids == expected_seq_ids
+
+
+@pytest.fixture
+def fake_sequence_group_metadata():
+    seq_ids = list(range(3))
+    return [
+        SequenceGroupMetadata(
+            request_id=str(i),
+            is_prompt=True,
+            seq_data={
+                i: MagicMock(),
+            },
+            sampling_params=MagicMock(),
+            block_tables={
+                i: MagicMock(),
+            },
+            lora_request=None,
+        ) for i in seq_ids
+    ]
+
+
+def test_filter_zero_length_proposals(fake_sequence_group_metadata):
+    proposal_lens = [0, 1, 0]
+    filtered_groups, indices = split_batch_by_proposal_len(
+        fake_sequence_group_metadata,
+        proposal_lens,
+        select_proposal_len_zero=True)
+
+    expected_groups = [
+        fake_sequence_group_metadata[0], fake_sequence_group_metadata[2]
+    ]
+    expected_indices = [0, 2]
+
+    assert filtered_groups == expected_groups
+    assert indices == expected_indices
+
+
+def test_filter_non_zero_length_proposals(fake_sequence_group_metadata):
+    proposal_lens = [0, 1, 2]
+    filtered_groups, indices = split_batch_by_proposal_len(
+        fake_sequence_group_metadata,
+        proposal_lens,
+        select_proposal_len_zero=False)
+
+    expected_groups = [
+        fake_sequence_group_metadata[1], fake_sequence_group_metadata[2]
+    ]
+    expected_indices = [1, 2]
+
+    assert filtered_groups == expected_groups
+    assert indices == expected_indices
+
+
+def test_empty_inputs():
+    filtered_groups, indices = split_batch_by_proposal_len(
+        [], [], select_proposal_len_zero=True)
+
+    assert filtered_groups == []
+    assert indices == []
+
+
+def test_all_zero_with_non_zero_filter(fake_sequence_group_metadata):
+    proposal_lens = [0, 0, 0]
+    filtered_groups, indices = split_batch_by_proposal_len(
+        fake_sequence_group_metadata,
+        proposal_lens,
+        select_proposal_len_zero=False)
+
+    assert filtered_groups == []
+    assert indices == []
+
+
+def test_all_non_zero_with_zero_filter(fake_sequence_group_metadata):
+    proposal_lens = [1, 1, 1]
+    filtered_groups, indices = split_batch_by_proposal_len(
+        fake_sequence_group_metadata,
+        proposal_lens,
+        select_proposal_len_zero=True)
+
+    assert filtered_groups == []
+    assert indices == []

tests/spec_decode/utils.py

@@ -1,13 +1,16 @@
 import torch
-from typing import List, Optional, Dict
+from typing import List, Optional, Dict, Iterable, Union
+from unittest.mock import MagicMock
 
 from vllm.worker.worker import Worker
 from vllm.utils import get_distributed_init_method, get_ip, get_open_port
 from vllm.engine.arg_utils import EngineArgs
-from vllm.sequence import Logprob, SequenceGroupMetadata, SequenceData
+from vllm.sequence import (Logprob, SequenceGroupMetadata, SequenceData,
+                           SamplerOutput, SequenceGroupOutput, SequenceOutput)
 from vllm.sampling_params import SamplingParams
 from vllm.worker.cache_engine import CacheEngine
 from vllm.model_executor.utils import set_random_seed
+from itertools import count
 from dataclasses import dataclass, fields
 
 
@@ -24,6 +27,11 @@ class ExecuteModelData:
         return dict(
             (field.name, getattr(self, field.name)) for field in fields(self))
 
+    @classmethod
+    def from_dict(cls, d):
+        cleaned = dict((field.name, d[field.name]) for field in fields(cls))
+        return cls(**cleaned)
+
 
 def round_up_to_next_block(seq_len: int, block_size: int) -> int:
     return (seq_len + block_size - 1) // block_size
@@ -50,6 +58,21 @@ def create_execute_model_data(
     )
 
 
+def mock_worker(cls=None,
+                vocab_size: int = 30_000,
+                max_model_len: int = 2048,
+                rank: int = 0) -> MagicMock:
+    if cls is None:
+        cls = Worker
+
+    worker = MagicMock(spec=cls)
+    worker.vocab_size = vocab_size
+    worker.max_model_len = max_model_len
+    worker.rank = rank
+    worker.device = 'cuda:0'
+    return worker
+
+
 def patch_execute_model_with_seeds(worker: Worker, rand_seeds: List[int]):
     seed_iter = iter(rand_seeds)
     original_execute_model = worker.execute_model
@@ -117,25 +140,12 @@ def create_seq_group_metadata_from_prompts(
     block_size: int,
     final_seq_lens: List[int],
     continuations: Optional[List[List[int]]] = None,
-    num_tokens_processed: Optional[List[int]] = None,
     seq_ids: Optional[List[int]] = None,
 ) -> List[SequenceGroupMetadata]:
 
     if continuations is None:
         continuations = [[] for _ in prompts]
 
-    if num_tokens_processed is None:
-        # Default to 1 token missing from kv cache for generation sequences.
-        num_tokens_processed = []
-        for continuation, prompt in zip(continuations, prompts):
-            # If prefill, then default to zero tokens processed.
-            if not continuation:
-                num_tokens_processed.append(0)
-            else:
-                # If generation, then default to all but one tokens processed.
-                num_tokens_processed.append(
-                    len(continuation) + len(prompt) - 1)
-
     if seq_ids is None:
         seq_ids = list(i for i, _ in enumerate(prompts))
 
@@ -155,13 +165,15 @@ def create_seq_group_metadata_from_prompts(
             is_prompt=len(cont_token_ids) == 0,
             seq_data={
                 i:
-                SequenceData(prompt_token_ids=prompt_token_ids[:] +
+                SequenceData(
-                             cont_token_ids[:])
+                    prompt_token_ids=prompt_token_ids[:],
+                    output_token_ids=cont_token_ids[:],
+                ),
             },
             sampling_params=SamplingParams(temperature=0.0, ),
             block_tables={i: block_allocations[i][:]},
-        ) for i, (prompt_token_ids, cont_token_ids, num_tokens_saved) in
+        ) for i, (prompt_token_ids,
-        enumerate(zip(prompts, continuations, num_tokens_processed))
+                  cont_token_ids) in enumerate(zip(prompts, continuations))
     ]
 
 
@@ -178,3 +190,68 @@ def assert_logprobs_dict_allclose(
             expected = torch.tensor(
                 single_step_expected_logprobs[token_id].logprob)
             assert torch.allclose(actual, expected)
+
+
+def create_sampler_output_list(
+        token_ids: torch.Tensor,
+        probs: Iterable[Optional[torch.Tensor]],
+        seq_ids: Optional[List[int]] = None) -> List[SamplerOutput]:
+    num_steps, batch_size = token_ids.shape
+    token_ids_by_step = token_ids.tolist()
+
+    if seq_ids is None:
+        seq_ids = list(range(batch_size))
+
+    return [
+        SamplerOutput(outputs=[
+            SequenceGroupOutput(
+                samples=[
+                    SequenceOutput(
+                        output_token=token_id,
+                        parent_seq_id=seq_ids[seq_index],
+                        logprobs={token_id: 0},
+                    )
+                ],
+                prompt_logprobs=None,
+            ) for seq_index, token_id in enumerate(token_ids_by_step[step])
+        ],
+                      sampled_token_probs=probs[step],
+                      sampled_token_ids=token_ids[step])
+        for step in range(num_steps)
+    ]
+
+
+def create_batch(batch_size,
+                 k,
+                 prompt_len: Union[int, List[int]] = 10,
+                 prev_output_token_len: int = 10,
+                 seq_ids: Optional[List[int]] = None,
+                 num_gpu_blocks: Optional[int] = None,
+                 block_size: Optional[int] = None):
+    if block_size is None:
+        block_size = 8
+
+    if num_gpu_blocks is None:
+        num_gpu_blocks = 2048 // block_size
+
+    iterator = count()
+
+    if isinstance(prompt_len, int):
+        prompt_lens = [prompt_len for _ in range(batch_size)]
+    else:
+        prompt_lens = prompt_len
+
+    prompts = [[next(iterator) for _ in range(p_len)] for p_len in prompt_lens]
+    prev_output_tokens = [[
+        next(iterator) for _ in range(prev_output_token_len)
+    ] for _ in range(batch_size)]
+    final_seq_lens = [
+        len(prompt) + len(prev_output_token) + k + 1
+        for prompt, prev_output_token in zip(prompts, prev_output_tokens)
+    ]
+
+    execute_model_data = create_execute_model_data(
+        create_seq_group_metadata_from_prompts(prompts, num_gpu_blocks,
+                                               block_size, final_seq_lens,
+                                               prev_output_tokens, seq_ids), )
+    return execute_model_data, prompts, prev_output_tokens

tests/test_sequence.py

@@ -0,0 +1,50 @@
+import pytest
+
+from vllm.sequence import SequenceGroupOutput, SamplerOutput, SequenceOutput
+
+
+@pytest.fixture
+def sample_outputs():
+    return [
+        SequenceGroupOutput(samples=[
+            SequenceOutput(parent_seq_id=0, output_token=i, logprobs={})
+        ],
+                            prompt_logprobs=None) for i in range(5)
+    ]
+
+
+@pytest.fixture
+def sampler_output(sample_outputs):
+    return SamplerOutput(outputs=sample_outputs)
+
+
+def test_sampler_output_initialization(sampler_output, sample_outputs):
+    assert len(sampler_output) == len(sample_outputs)
+    assert sampler_output.sampled_token_probs is None
+    assert sampler_output.sampled_token_ids is None
+    assert sampler_output.spec_decode_worker_metrics is None
+
+
+def test_sampler_output_getitem(sampler_output, sample_outputs):
+    assert sampler_output[2] == sample_outputs[2]
+
+
+def test_sampler_output_setitem(sampler_output):
+    new_output = SequenceGroupOutput(samples=[
+        SequenceOutput(parent_seq_id=0, output_token=99, logprobs={})
+    ],
+                                     prompt_logprobs=None)
+    sampler_output[2] = new_output
+    assert sampler_output[2] == new_output
+
+
+def test_sampler_output_len(sampler_output, sample_outputs):
+    assert len(sampler_output) == len(sample_outputs)
+
+
+def test_sampler_output_eq(sample_outputs):
+    sampler_output1 = SamplerOutput(outputs=sample_outputs)
+    sampler_output2 = SamplerOutput(outputs=sample_outputs.copy())
+    sampler_output3 = SamplerOutput(outputs=sample_outputs[:-1])
+    assert sampler_output1 == sampler_output2
+    assert sampler_output1 != sampler_output3

vllm/model_executor/layers/rejection_sampler.py

@@ -21,8 +21,6 @@ class RejectionSampler(nn.Module):
                 nontrivial latency.
         """
         super().__init__()
-        self.probs_dtype = torch.float32
-        self.token_id_dtype = torch.int64
         self._strict_mode = strict_mode
 
         # NOTE: A "bonus token" is accepted iff all proposal tokens are
@@ -44,6 +42,14 @@ class RejectionSampler(nn.Module):
                                                dtype=torch.long,
                                                device=device)
 
+    @property
+    def probs_dtype(self):
+        return torch.float32
+
+    @property
+    def token_id_dtype(self):
+        return torch.int64
+
     def forward(
         self,
         target_probs: torch.Tensor,

vllm/model_executor/layers/sampler.py

@@ -587,4 +587,4 @@ def _build_sampler_output(
                 SequenceOutput(seq_ids[parent_id], next_token_id, logprobs))
         sampler_output.append(
             SequenceGroupOutput(seq_outputs, group_prompt_logprobs))
-    return sampler_output
+    return SamplerOutput(outputs=sampler_output)

vllm/sequence.py

@@ -2,12 +2,16 @@
 import copy
 import enum
 from dataclasses import dataclass
-from typing import Dict, List, Optional, Union
+from typing import Dict, List, Optional, Union, TYPE_CHECKING
 
 from vllm.block import LogicalTokenBlock
 from vllm.sampling_params import SamplingParams
 from vllm.lora.request import LoRARequest
 
+if TYPE_CHECKING:
+    import torch
+    from vllm.spec_decode.metrics import SpecDecodeWorkerMetrics
+
 
 @dataclass
 class Logprob:
@@ -81,6 +85,8 @@ class SequenceData:
 
     Args:
         prompt_token_ids: The token IDs of the prompt.
+        output_token_ids: The token IDs of the output. Set to an empty list if
+            None.
 
     Attributes:
         prompt_token_ids: The token IDs of the prompt.
@@ -91,9 +97,13 @@ class SequenceData:
     def __init__(
         self,
         prompt_token_ids: List[int],
+        output_token_ids: Optional[List[int]] = None,
     ) -> None:
+        if output_token_ids is None:
+            output_token_ids = []
+
         self.prompt_token_ids = prompt_token_ids
-        self.output_token_ids: List[int] = []
+        self.output_token_ids = output_token_ids
         self.cumulative_logprob = 0.0
 
     def append_token_id(self, token_id: int, logprob: float) -> None:
@@ -117,6 +127,12 @@ class SequenceData:
             return self.prompt_token_ids[-1]
         return self.output_token_ids[-1]
 
+    def get_prompt_token_ids(self) -> int:
+        return self.prompt_token_ids
+
+    def get_output_token_ids(self) -> int:
+        return self.output_token_ids
+
     def __repr__(self) -> str:
         return (f"SequenceData("
                 f"prompt_token_ids={self.prompt_token_ids}, "
@@ -506,6 +522,35 @@ class SequenceGroupOutput:
                 and self.prompt_logprobs == other.prompt_logprobs)
 
 
-# For each sequence group, we generate a list of SequenceOutput object,
+@dataclass
-# each of which contains one possible candidate for the next token.
+class SamplerOutput:
-SamplerOutput = List[SequenceGroupOutput]
+    """For each sequence group, we generate a list of SequenceOutput object,
+    each of which contains one possible candidate for the next token.
+
+    This datastructure implements methods so it can be used like a list, but
+    also has optional fields for device tensors.
+    """
+
+    outputs: List[SequenceGroupOutput]
+
+    # On-device tensor containing probabilities of each token.
+    sampled_token_probs: Optional["torch.Tensor"] = None
+
+    # On-device tensor containing the sampled token ids.
+    sampled_token_ids: Optional["torch.Tensor"] = None
+
+    # Spec decode metrics populated by workers.
+    spec_decode_worker_metrics: Optional["SpecDecodeWorkerMetrics"] = None
+
+    def __getitem__(self, idx: int):
+        return self.outputs[idx]
+
+    def __setitem__(self, idx: int, value):
+        self.outputs[idx] = value
+
+    def __len__(self):
+        return len(self.outputs)
+
+    def __eq__(self, other: object):
+        return isinstance(other,
+                          self.__class__) and self.outputs == other.outputs

vllm/spec_decode/batch_expansion.py

@@ -0,0 +1,351 @@
+from typing import Iterator, List, Tuple, Optional, Dict
+from itertools import chain, count
+
+import torch
+
+from vllm.sequence import (SamplerOutput, SequenceGroupMetadata, SequenceData)
+from vllm.worker.worker import Worker
+from vllm.spec_decode.util import nvtx_range, sampler_output_to_torch, get_all_seq_ids, split_batch_by_proposal_len
+from vllm.spec_decode.interfaces import SpeculativeScorer, SpeculativeProposals, SpeculativeScores
+
+SeqId = int
+TargetSeqId = int
+TokenId = int
+
+
+class BatchExpansionTop1Scorer(SpeculativeScorer):
+    """Implements a speculative scorer that uses batch expansion to get
+    probabilities of speculative tokens according to the scoring model.
+
+    Batch expansion converts a list of sequences and multiple query positions
+    to a new batch of sequences, each with a single query position. This allows
+    for MQA-like scoring in speculative decoding without requiring an MQA
+    kernel.
+
+    It is strictly less efficient than MQA scoring.
+
+    It only supports scoring the top1 proposal tokens of the proposer, instead
+    of topk/tree.
+    """
+
+    def __init__(self, scorer_worker: Worker, device: str, vocab_size: int):
+        self._scorer_worker = scorer_worker
+        self._device = device
+        self._vocab_size = vocab_size
+
+    @nvtx_range("BatchExpansionTop1Scorer.score_proposals")
+    def score_proposals(
+        self,
+        seq_group_metadata_list: List[SequenceGroupMetadata],
+        blocks_to_swap_in: Optional[Dict[int, int]],
+        blocks_to_swap_out: Optional[Dict[int, int]],
+        blocks_to_copy: Optional[Dict[int, List[int]]],
+        k: int,
+        proposals: SpeculativeProposals,
+    ) -> SpeculativeScores:
+        """Score the proposed tokens via the scorer model.
+
+        This converts each input sequence to a set of k+1 target sequences. The
+        target sequences have the unique continuations to be scored and a
+        unique sequence ID that is different from all input sequence ids.
+
+        If a speculative sequence length would exceed the max model length, then
+        no speculation is produced for that sequence.
+
+        Args:
+            seq_group_metadata_list: The input sequence group metadata.
+            blocks_to_swap_in: This is passed to the worker during scoring.
+            blocks_to_swap_out: This is passed to the worker during scoring.
+            blocks_to_copy: This is passed to the worker during scoring.
+            k: The fixed proposal length.
+            proposals: The speculative proposals to score.
+        Returns:
+            SpeculativeScores: The scores of each speculative token, along with
+                which sequences were ignored during scoring.
+        """
+
+        # TODO(cade) perform this on GPU to remove blocking call.
+        proposal_lens_list = proposals.proposal_lens.tolist()
+        proposal_token_ids_list = proposals.proposal_token_ids.tolist()
+
+        spec_indices, non_spec_indices, target_seq_group_metadata_list, num_scoring_tokens = self._expand_batch(
+            seq_group_metadata_list=seq_group_metadata_list,
+            proposal_token_ids_list=proposal_token_ids_list,
+            proposal_lens_list=proposal_lens_list,
+        )
+
+        target_sampler_output = self._scorer_worker.execute_model(
+            seq_group_metadata_list=target_seq_group_metadata_list,
+            blocks_to_swap_in=blocks_to_swap_in,
+            blocks_to_swap_out=blocks_to_swap_out,
+            blocks_to_copy=blocks_to_copy,
+            return_python_output=False)
+
+        all_tokens, all_probs = self._contract_batch(
+            original_bs=len(seq_group_metadata_list),
+            target_sampler_output=target_sampler_output,
+            proposals=proposals,
+            num_scoring_tokens=num_scoring_tokens,
+            non_spec_indices=non_spec_indices,
+            spec_indices=spec_indices,
+            k=k,
+        )
+
+        return SpeculativeScores(
+            probs=all_probs,
+            token_ids=all_tokens,
+        )
+
+    def _expand_batch(
+        self,
+        seq_group_metadata_list: List[SequenceGroupMetadata],
+        proposal_token_ids_list: List[TokenId],
+        proposal_lens_list: List[int],
+    ) -> Tuple[List[int], List[int], List[SequenceGroupMetadata], int]:
+        """Given the input sequences and potentially multiple corresponding
+        proposal tokens, create a new batch where each sequence has a single
+        query token.
+        """
+
+        # vLLM currently only supports proposal lens equal to zero or the batch
+        # proposal len. This adds some complexity (splitting the batch into spec
+        # and non spec sequences) and should be removed in the future. It can be
+        # done by supporting per-sequence proposal lens.
+        spec_seqs, spec_indices = split_batch_by_proposal_len(
+            seq_group_metadata_list,
+            proposal_lens_list,
+            select_proposal_len_zero=False)
+        non_spec_seqs, non_spec_indices = split_batch_by_proposal_len(
+            seq_group_metadata_list,
+            proposal_lens_list,
+            select_proposal_len_zero=True)
+
+        target_seq_group_metadata_list = self._create_scoring_model_input(
+            spec_seqs, proposal_token_ids_list)
+        num_scoring_tokens = len(target_seq_group_metadata_list)
+        target_seq_group_metadata_list.extend(non_spec_seqs)
+
+        return spec_indices, non_spec_indices, target_seq_group_metadata_list, num_scoring_tokens
+
+    def _contract_batch(self, original_bs: int,
+                        target_sampler_output: List[SamplerOutput],
+                        proposals: SpeculativeProposals,
+                        num_scoring_tokens: int, non_spec_indices: List[int],
+                        spec_indices: List[int],
+                        k: int) -> Tuple[torch.Tensor, torch.Tensor]:
+        """Contract the expanded batch back into its original size.
+        This maps the scores of speculative tokens back to their original
+        sequences.
+        """
+        (target_token_ids, target_probs, non_spec_target_token_ids,
+         non_spec_target_probs) = self._split_scoring_output(
+             target_sampler_output, num_scoring_tokens)
+
+        # Map distinct sequences used to score each token
+        # of shape [batch_size * k + 1] back to [batch_size, k + 1].
+        batch_size, k = proposals.proposal_token_ids.shape
+
+        target_token_ids = target_token_ids.squeeze().reshape(
+            batch_size, k + 1)
+        target_probs = target_probs.squeeze().reshape(batch_size, k + 1,
+                                                      self._vocab_size)
+
+        all_tokens = torch.full(size=(original_bs, k + 1),
+                                fill_value=-1,
+                                device=self._device,
+                                dtype=torch.long)
+        all_probs = torch.zeros(original_bs,
+                                k + 1,
+                                self._vocab_size,
+                                device=self._device,
+                                dtype=torch.float32)
+
+        if non_spec_indices:
+            all_tokens[non_spec_indices, 0] = non_spec_target_token_ids
+            all_probs[non_spec_indices, :1, :] = non_spec_target_probs
+
+        if spec_indices:
+            all_tokens[spec_indices] = target_token_ids
+            all_probs[spec_indices] = target_probs
+
+        return all_tokens, all_probs
+
+    def _create_scoring_model_input(
+            self,
+            seq_group_metadata_list: List[SequenceGroupMetadata],
+            proposal_token_ids: List[List[TokenId]],  # shape: [batch_size, k]
+    ) -> List[SequenceGroupMetadata]:
+        """Given the original input sequences and proposed tokens from the draft
+        model, create a list of target sequences that can be used for scoring.
+        """
+
+        if not seq_group_metadata_list:
+            return []
+
+        target_seq_ids_iter = self._create_target_seq_id_iterator(
+            get_all_seq_ids(seq_group_metadata_list))
+
+        target_seq_group_metadata = list(
+            chain.from_iterable(
+                self._create_target_seq_group_metadata(
+                    seq_group_metadata,
+                    proposal_token_ids,
+                    i,
+                    target_seq_ids_iter,
+                ) for i, seq_group_metadata in enumerate(
+                    seq_group_metadata_list)))
+
+        return target_seq_group_metadata
+
+    def _create_target_seq_group_metadata(
+        self,
+        input_seq_group_metadata: SequenceGroupMetadata,
+        proposal_token_ids: List[TokenId],  # shape: [batch_size, k]
+        batch_index: int,
+        target_seq_ids_iter: Iterator[TargetSeqId],
+    ) -> List[SequenceGroupMetadata]:
+        """Given an input sequence group metadata and a list of draft tokens,
+        create a list of target SequenceGroupMetadata, one for each
+        token id that needs to be scored.
+
+        Naive speculative decoding requires K target model scores, one for each
+        draft model token. However one can add a bonus token such that if each
+        token is accepted, then a final token may be sampled from the model.
+        This function creates K+1 target SequenceGroupMetadata to take
+        advantage of the bonus token.
+        """
+        assert not input_seq_group_metadata.is_prompt, (
+            "Speculating on "
+            "prompts not yet supported")
+        assert len(input_seq_group_metadata.seq_data) == 1, (
+            "Beam search "
+            "not supported in speculative decoding")
+        input_seq_id = next(iter(input_seq_group_metadata.seq_data.keys()))
+
+        token_ids_to_score = self._get_token_ids_to_score(
+            proposal_token_ids[batch_index])
+
+        target_seq_group_metadata_list: List[SequenceGroupMetadata] = []
+        for token_ids in token_ids_to_score:
+            target_seq_group_metadata_list.append(
+                self._create_single_target_seq_group_metadata(
+                    input_seq_group_metadata,
+                    input_seq_id,
+                    next(target_seq_ids_iter),
+                    token_ids,
+                ))
+
+        return target_seq_group_metadata_list
+
+    def _create_single_target_seq_group_metadata(
+        self,
+        seq_group_metadata: SequenceGroupMetadata,
+        seq_id: SeqId,
+        target_seq_id: TargetSeqId,
+        token_ids: List[TokenId],
+    ) -> SequenceGroupMetadata:
+        """Create a single target SequenceGroupMetadata.
+
+        Args:
+            seq_group_metadata: The metadata for the input sequence.
+            seq_id: The input sequence ID.
+            target_seq_id: The corresponding target sequence ID.
+            token_ids: The list of token ids that are to be appended to the
+                input sequence.
+        """
+        seq_data = seq_group_metadata.seq_data[seq_id]
+        prompt_token_ids = seq_data.get_prompt_token_ids()
+        new_output_token_ids = [*seq_data.get_output_token_ids(), *token_ids]
+
+        return SequenceGroupMetadata(
+            request_id=seq_group_metadata.request_id,
+            is_prompt=seq_group_metadata.is_prompt,
+            seq_data={
+                target_seq_id:
+                SequenceData(
+                    prompt_token_ids=prompt_token_ids,
+                    output_token_ids=new_output_token_ids,
+                ),
+            },
+            sampling_params=seq_group_metadata.sampling_params,
+            block_tables={
+                target_seq_id: seq_group_metadata.block_tables[seq_id],
+            },
+            lora_request=None,
+        )
+
+    def _split_scoring_output(
+        self, sampler_output: SamplerOutput, num_scoring_tokens: int
+    ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]:
+        """Split the target model output into speculative and non-speculative
+        output.
+        """
+
+        # vLLM currently only supports proposal lens equal to zero or the batch
+        # proposal len. This adds some complexity (splitting the batch into spec
+        # and non spec sequences) and should be removed in the future. It can be
+        # done by supporting per-sequence proposal lens.
+        #
+        # First samples are from speculative scoring, latter samples are non-
+        # speculative samples.
+        split_sizes = [
+            num_scoring_tokens,
+            sampler_output.sampled_token_ids.numel() - num_scoring_tokens
+        ]
+        (spec_probs, non_spec_probs
+         ) = sampler_output.sampled_token_probs.split(split_sizes)
+        (spec_sampled_tokens, non_spec_sampled_tokens
+         ) = sampler_output.sampled_token_ids.flatten().split(split_sizes)
+
+        # Convert scores to tensors.
+        sampler_output.sampled_token_probs = spec_probs
+        sampler_output.sampled_token_ids = spec_sampled_tokens
+        target_token_ids, target_probs = sampler_output_to_torch(
+            [sampler_output])
+
+        # Convert non-speculative output tokens to tensors.
+        sampler_output.sampled_token_probs = non_spec_probs
+        sampler_output.sampled_token_ids = non_spec_sampled_tokens
+        non_spec_target_token_ids, non_spec_target_probs = sampler_output_to_torch(
+            [sampler_output])
+
+        return target_token_ids, target_probs, non_spec_target_token_ids, non_spec_target_probs
+
+    def _create_target_seq_id_iterator(
+            self, seq_ids: List[SeqId]) -> Iterator[TargetSeqId]:
+        """Create an iterator for creating target sequence ids.
+        Target sequence ids are distinct from sequence ids because we create a
+        distinct target sequence id for each proposal token to be scored.
+
+        This implementation increments a counter starting at 1 + max of all
+        provided input sequence ids.
+        """
+        return count(start=max(seq_ids) + 1)
+
+    def _get_token_ids_to_score(
+        self,
+        full_spec_token_ids: List[TokenId]  # shape: [k]
+    ) -> List[List[TokenId]]:
+        """Given an int tensor of proposal token ids, return a list of
+        token ids that should be scored.
+
+        Returns k+1 output lists. The additional one is used for generating the
+        bonus token.
+
+        Example:
+            Input: [0, 1, 2, 3] (k=4)
+            Output: (k+1 lists)
+                []
+                [0]
+                [0, 1]
+                [0, 1, 2]
+                [0, 1, 2, 3]
+        """
+        empty_token_ids = []
+
+        token_ids_to_score = [empty_token_ids]
+        token_ids_to_score.extend([
+            full_spec_token_ids[:i + 1]
+            for i in range(len(full_spec_token_ids))
+        ])
+        return token_ids_to_score

vllm/spec_decode/interfaces.py

@@ -0,0 +1,77 @@
+from typing import List, Tuple, Optional, Dict
+from dataclasses import dataclass
+from abc import ABC, abstractmethod
+
+import torch
+
+from vllm.sequence import SequenceGroupMetadata
+
+
+@dataclass
+class SpeculativeProposals:
+    """Datastructure used to represent proposal tokens from some proposer. It
+    also tracks how many speculative tokens each sequence has.
+    """
+
+    # Speculative proposal tokens.
+    proposal_token_ids: torch.Tensor
+
+    # Probabilities of the proposal tokens according to the proposer.
+    proposal_probs: torch.Tensor
+
+    # The valid length of each proposal; can be zero.
+    proposal_lens: torch.Tensor
+
+    def __repr__(self):
+        return (f"SpeculativeProposals("
+                f"proposal_token_ids={self.proposal_token_ids.shape}, "
+                f"proposal_probs={self.proposal_probs.shape}, "
+                f"proposal_lens={self.proposal_lens.shape})")
+
+
+@dataclass
+class SpeculativeScores:
+    """Datastructure used to represent the scores of speculative tokens
+    according to the scoring model.
+    """
+
+    # Probabilities of the speculative tokens according to the scoring model.
+    probs: torch.Tensor
+
+    # Token ids sampled from the scoring model. Used for speculative bonus
+    # tokens and also non-speculative normal decoding.
+    token_ids: torch.Tensor
+
+    def __repr__(self):
+        return (f"SpeculativeScores("
+                f"probs={self.probs.shape}, "
+                f"token_ids={self.token_ids.shape})")
+
+
+class SpeculativeProposer(ABC):
+
+    @abstractmethod
+    def get_proposals(
+        self,
+        seq_group_metadata_list: List[SequenceGroupMetadata],
+        blocks_to_swap_in: Dict[int, int],
+        blocks_to_swap_out: Dict[int, int],
+        blocks_to_copy: Dict[int, List[int]],
+        max_proposal_len: int,
+    ) -> SpeculativeProposals:
+        raise NotImplementedError
+
+
+class SpeculativeScorer(ABC):
+
+    @abstractmethod
+    def score_proposals(
+        self,
+        seq_group_metadata_list: List[SequenceGroupMetadata],
+        blocks_to_swap_in: Optional[Dict[int, int]],
+        blocks_to_swap_out: Optional[Dict[int, int]],
+        blocks_to_copy: Optional[Dict[int, List[int]]],
+        k: int,
+        proposals: SpeculativeProposals,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        raise NotImplementedError

vllm/spec_decode/metrics.py

@@ -0,0 +1,174 @@
+import torch
+from dataclasses import dataclass
+from vllm.model_executor.layers.rejection_sampler import RejectionSampler
+from typing import Optional
+from vllm.utils import in_wsl
+import time
+from typing import Callable
+
+
+@dataclass
+class SpecDecodeWorkerMetrics:
+    """Dataclass holding metrics emitted from the spec decode worker.
+    """
+
+    # The empirical acceptance rate of the proposal method on a per-token basis.
+    # This is useful for evaluating how well the proposal method aligns with the
+    # scoring method.
+    draft_acceptance_rate: float
+
+    # The empirical efficiency, measured as the number of tokens emitted by the
+    # system divided by the number of tokens that could be emitted by the system
+    # if the proposal method were perfect.
+    system_efficiency: float
+
+    # The number of speculative tokens produced by the proposal method.
+    draft_tokens: int
+
+    # The number of tokens emitted by the entire system.
+    emitted_tokens: int
+
+    # The number of tokens accepted by the scoring model and verification
+    # routine, e.g. Llama2-70B and lossless rejection sampling.
+    #
+    # NOTE: Any token accepted by the verification routine is considered
+    # accepted (regardless of if the speculative prefix is also accepted). The
+    # user will usually see less accepted tokens. This metric is helpful when
+    # evaluating alignment of the proposal method with the scoring model.
+    accepted_tokens: int
+
+    # The number of speculative tokens per sequence.
+    num_spec_tokens: int
+
+
+Timer = Callable[[], float]
+
+
+class AsyncMetricsCollector:
+    """Class which copies rejection sampler metrics from the device to CPU on a
+    non-default Torch stream.
+    """
+
+    def __init__(self,
+                 rejection_sampler: RejectionSampler,
+                 timer: Optional[Timer] = None,
+                 collect_interval_s: float = 5.0):
+        self._rejection_sampler = rejection_sampler
+        self._timer = time.time if timer is None else timer
+
+        self._rank: Optional[int] = None
+
+        # We don't have a device set yet.
+        self._copy_stream: Optional[torch.cuda.Stream] = None
+
+        self._in_flight_copy: Optional[torch.cuda.Event] = None
+
+        pin_memory = not in_wsl()
+        self._aggregate_num_accepted_tokens = torch.tensor(
+            0, dtype=torch.long, device="cpu", pin_memory=pin_memory)
+        self._aggregate_num_emitted_tokens = torch.tensor(
+            0, dtype=torch.long, device="cpu", pin_memory=pin_memory)
+        self._aggregate_num_draft_tokens = 0
+
+        self._rejsample_metrics_collect_interval_s = collect_interval_s
+        self._last_metrics_collect_time = self._timer()
+
+    def init_gpu_tensors(self, rank: int) -> None:
+        self._rank = rank
+        self._copy_stream = torch.cuda.Stream()
+
+    def maybe_collect_rejsample_metrics(
+            self, k: int) -> Optional[SpecDecodeWorkerMetrics]:
+
+        # If a copy was initiated in the previous call, collect and return.
+        if self._in_flight_copy is not None:
+            ready_event = self._in_flight_copy
+            self._in_flight_copy = None
+            return self._collect_rejsample_metrics(k, ready_event)
+
+        # Otherwise, check if we should start a new copy.
+        if self._should_collect_rejsample_metrics(self._timer()):
+            assert self._in_flight_copy is None
+            self._in_flight_copy = self._copy_rejsample_metrics_async()
+
+        return None
+
+    def _should_collect_rejsample_metrics(self, now: float) -> bool:
+        """Return whether or not this iteration should print rejection sampling
+        metrics.
+        """
+        if self._rank != 0:
+            return False
+
+        if (now - self._last_metrics_collect_time <
+                self._rejsample_metrics_collect_interval_s):
+            return False
+        return True
+
+    def _copy_rejsample_metrics_async(self) -> torch.cuda.Event:
+        """Copy rejection sampling metrics (number of accepted tokens, etc) to
+        CPU asynchronously.
+
+        Returns a CUDA event recording when the copy is complete.
+        """
+        self._copy_stream.wait_stream(torch.cuda.current_stream())
+
+        with torch.cuda.stream(self._copy_stream):
+            self._aggregate_num_accepted_tokens.copy_(
+                self._rejection_sampler.num_accepted_tokens, non_blocking=True)
+            self._aggregate_num_emitted_tokens.copy_(
+                self._rejection_sampler.num_emitted_tokens, non_blocking=True)
+            # Number of draft tokens is calculated on CPU, so no copy is
+            # required.
+            self._aggregate_num_draft_tokens = (
+                self._rejection_sampler.num_draft_tokens)
+
+        aggregate_metrics_ready = torch.cuda.Event()
+        aggregate_metrics_ready.record(self._copy_stream)
+
+        return aggregate_metrics_ready
+
+    def _collect_rejsample_metrics(
+            self, k: int,
+            ready_event: torch.cuda.Event) -> SpecDecodeWorkerMetrics:
+        """Create metrics object from statistics copied asynchronously.
+
+        Args:
+            k: int. The number of speculative tokens; used to determine system
+                efficiency.
+            ready_event: torch.cuda.Event. The CUDA event recording when the
+                async GPU->CPU copy is complete.
+        """
+
+        ready_event.synchronize()
+        accepted_tokens = self._aggregate_num_accepted_tokens.item()
+        emitted_tokens = self._aggregate_num_emitted_tokens.item()
+        draft_tokens = self._aggregate_num_draft_tokens
+
+        num_possible_tokens = self.get_max_num_accepted_tokens(draft_tokens, k)
+
+        if draft_tokens > 0:
+            draft_acceptance_rate = accepted_tokens / draft_tokens
+        else:
+            draft_acceptance_rate = float("nan")
+
+        if num_possible_tokens > 0:
+            system_efficiency = emitted_tokens / num_possible_tokens
+        else:
+            system_efficiency = float("nan")
+
+        return SpecDecodeWorkerMetrics(
+            num_spec_tokens=k,
+            draft_acceptance_rate=draft_acceptance_rate,
+            system_efficiency=system_efficiency,
+            accepted_tokens=accepted_tokens,
+            draft_tokens=draft_tokens,
+            emitted_tokens=emitted_tokens,
+        )
+
+    @staticmethod
+    def get_max_num_accepted_tokens(draft_tokens: int, k: int) -> int:
+        # Divide by k since batch size can be variable.
+        total_num_spec_seqs = draft_tokens / k
+        num_accepted_per_seq_if_all_accepted = k + 1
+        return int(total_num_spec_seqs / num_accepted_per_seq_if_all_accepted)

vllm/spec_decode/multi_step_worker.py

@@ -0,0 +1,366 @@
+from typing import List, Dict, Optional, Tuple
+import copy
+
+import torch
+
+from vllm.sequence import SamplerOutput, SequenceGroupMetadata
+from vllm.worker.worker import Worker
+from vllm.spec_decode.interfaces import SpeculativeProposals, SpeculativeProposer
+from vllm.spec_decode.util import sampler_output_to_torch
+
+
+class MultiStepWorker(Worker):
+    """The MultiStepWorker is equivalent to a Worker except that it allows
+    multiple forward passes in a single call, assuming the scheduler has
+    allocated enough space to store the additional KV. This reduces overhead
+    by invoking the scheduler less.
+
+    The MultiStepWorker does not support cache swap operations, or beam search.
+    Cache swap operations do not require large modifications. On the other hand,
+    beam search requires memory allocations during sequence forks and thus
+    requires more thought for MultiStepWorker support.
+    """
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+        self._proposer: Optional[DraftModelTop1Proposer] = None
+
+    def init_model(self):
+        super().init_model()
+
+        self._proposer = DraftModelTop1Proposer(
+            self,
+            self.device,
+            self.max_model_len,
+            self.vocab_size,
+        )
+
+    @torch.inference_mode()
+    def execute_model_multi_step(
+        self,
+        seq_group_metadata_list: List[SequenceGroupMetadata],
+        blocks_to_swap_in: Dict[int, int],
+        blocks_to_swap_out: Dict[int, int],
+        blocks_to_copy: Dict[int, List[int]],
+        num_steps: int,
+    ) -> List[SamplerOutput]:
+        """Run the model forward pass num_steps times. Returns the list of
+        sampler output, one per model forward pass.
+        """
+        self._raise_if_unsupported(seq_group_metadata_list, blocks_to_swap_in,
+                                   blocks_to_swap_out, blocks_to_copy)
+
+        # Shallow copy input data so modifications (such as appending tokens)
+        # do not cause side-effects.
+        copied_seq_group_metadata_list = self._shallow_copy_inputs(
+            seq_group_metadata_list)
+
+        # Assert enough KV space for num_steps tokens per sequence.
+        self._assert_enough_kv_space(seq_group_metadata_list, num_steps)
+
+        # Run model num_steps times.
+        model_outputs = []
+        for _ in range(num_steps):
+            model_output = super().execute_model(
+                seq_group_metadata_list=copied_seq_group_metadata_list,
+                blocks_to_swap_in=blocks_to_swap_in,
+                blocks_to_swap_out=blocks_to_swap_out,
+                blocks_to_copy=blocks_to_copy,
+            )
+
+            self._append_new_tokens(model_output,
+                                    copied_seq_group_metadata_list)
+            model_outputs.append(model_output)
+
+        return model_outputs
+
+    def get_spec_proposals(
+        self,
+        seq_group_metadata_list: List[SequenceGroupMetadata],
+        blocks_to_swap_in: Dict[int, int],
+        blocks_to_swap_out: Dict[int, int],
+        blocks_to_copy: Dict[int, List[int]],
+        max_proposal_len: int,
+    ) -> SpeculativeProposals:
+        """Produce speculations given an input batch of sequences. The number of
+        speculative tokens per sequence is determined by max_proposal_len.
+        """
+
+        return self._proposer.get_proposals(
+            seq_group_metadata_list,
+            blocks_to_swap_in,
+            blocks_to_swap_out,
+            blocks_to_copy,
+            max_proposal_len,
+        )
+
+    def _append_new_tokens(
+            self, model_output: SamplerOutput,
+            seq_group_metadata_list: SequenceGroupMetadata) -> None:
+        """Given model output from a single run, append the tokens to the
+        sequences. This is normally done outside of the worker, but it is
+        required if the worker is to perform multiple forward passes.
+        """
+        for seq_group_metadata, sequence_group_outputs in zip(
+                seq_group_metadata_list, model_output):
+            seq_group_metadata.is_prompt = False
+
+            for seq_output in sequence_group_outputs.samples:
+                # NOTE: Beam search is not supported, so we can assume that
+                # parent_seq_id == seq_id.
+                seq = seq_group_metadata.seq_data[seq_output.parent_seq_id]
+
+                token_id = seq_output.output_token
+                token_logprob = seq_output.logprobs[token_id]
+
+                seq.append_token_id(token_id, token_logprob.logprob)
+
+    def _shallow_copy_inputs(
+        self, seq_group_metadata_list: List[SequenceGroupMetadata]
+    ) -> List[SequenceGroupMetadata]:
+        """Copy input data structures to remove side-effects when input data
+        structures are shared with other modules.
+
+        Helpful when the vLLM scheduler runs in the same process as the worker.
+        The alternative is deep-copying (or other form of deep copy); this has
+        performance downsides.
+        """
+
+        # Shallow-copy the list of SequenceGroupMetadata. This allows us to
+        # append tokens and change is_prompt without external side-effects.
+        new_seq_group_metadata_list = []
+
+        for old_seq_group_metadata in seq_group_metadata_list:
+            # We must shallow-copy seq_group_metadata as is_prompt could change.
+            seq_group_metadata = copy.copy(old_seq_group_metadata)
+            new_seq_group_metadata_list.append(seq_group_metadata)
+
+            # We must shallow-copy seq_data as we will append token ids
+            new_seq_data = {}
+            for seq_id, old_seq_data in seq_group_metadata.seq_data.items():
+                new_seq_data[seq_id] = copy.copy(old_seq_data)
+                new_seq_data[
+                    seq_id].output_token_ids = old_seq_data.output_token_ids[:]
+
+            seq_group_metadata.seq_data = new_seq_data
+
+        return new_seq_group_metadata_list
+
+    def _assert_enough_kv_space(
+            self, seq_group_metadata_list: List[SequenceGroupMetadata],
+            num_steps: int) -> None:
+        """Assert there are enough physical blocks per sequence to store the
+        current KV plus additional KV from num_steps tokens.
+        """
+        assert self.model_runner.block_size is not None
+        for seq_group_metadata in seq_group_metadata_list:
+            # Only one seq_id is guaranteed because there is no beam search.
+            seq_id = list(seq_group_metadata.seq_data.keys())[0]
+            seq = seq_group_metadata.seq_data[seq_id]
+
+            # After num_steps, the seq len will be the current seq len
+            # plus one token per step.
+            final_seq_len = seq.get_len() + num_steps
+
+            # We will have final_seq_len - 1 KV because vLLM saves KV for a
+            # token in the iteration after the token was generated.
+            required_num_kv_slots = final_seq_len - 1
+
+            # The allocated number of kv slots is the number of allocated blocks
+            # times the number of slots of block.
+            number_physical_blocks = len(
+                seq_group_metadata.block_tables[seq_id])
+            allocated_kv_slots = (number_physical_blocks *
+                                  self.model_runner.block_size)
+
+            if required_num_kv_slots > allocated_kv_slots:
+                request_id = seq_group_metadata.request_id
+                raise ValueError(
+                    "The worker attempted to run "
+                    f"{num_steps} times but found insufficient KV space for "
+                    f"{request_id=} {seq_id=}. ({allocated_kv_slots=} "
+                    f"{required_num_kv_slots=}).")
+
+    def _raise_if_unsupported(
+        self,
+        seq_group_metadata_list: List[SequenceGroupMetadata],
+        blocks_to_swap_in: Dict[int, int],
+        blocks_to_swap_out: Dict[int, int],
+        blocks_to_copy: Dict[int, List[int]],
+    ) -> None:
+        """MultiStepWorker does not yet implement support for cache swap
+        operations or beam search.
+        """
+        if any([blocks_to_swap_in, blocks_to_swap_out, blocks_to_copy]):
+            raise NotImplementedError(
+                "MultiStepWorker does not support cache operations")
+
+        if any(
+                len(seq_group_metadata.seq_data.keys()) != 1
+                for seq_group_metadata in seq_group_metadata_list):
+            raise NotImplementedError(
+                "MultiStepWorker does not support beam search.")
+
+
+class DraftModelTop1Proposer(SpeculativeProposer):
+    """Helper class which separates out sequences which would exceed the max
+    model length when speculated upon.
+
+    This allows combinations of models such as JackFram/llama-68m draft with
+    meta-llama/Llama2-13b-chat-hf, as llama-68m has max_position_embeddings of
+    2048 while Llama2-13b has max_position_embeddings of 4096.
+
+    We treat the sequences which exceed the proposal draft model length as
+    "non-spec sequences". Essentially they skip the draft model and go through
+    normal decoding in the target model.
+
+    Currently, only proposal_lens of 0 and k are supported, where k is a global
+    batch proposal length. In the future vLLM should support per-sequence
+    proposal lengths.
+    """
+
+    def __init__(
+        self,
+        draft_worker: MultiStepWorker,
+        device: str,
+        max_model_len: int,
+        vocab_size: int,
+    ):
+        self._draft_worker = draft_worker
+        self._device = device
+        self._max_model_len = max_model_len
+        self._vocab_size = vocab_size
+
+    def get_proposals(
+        self,
+        seq_group_metadata_list: List[SequenceGroupMetadata],
+        blocks_to_swap_in: Dict[int, int],
+        blocks_to_swap_out: Dict[int, int],
+        blocks_to_copy: Dict[int, List[int]],
+        max_proposal_len: int,
+    ) -> SpeculativeProposals:
+        """Get speculative proposals given the input batch.
+
+        Sequences which would exceed the max model length are skipped during
+        speculation.
+        """
+
+        # Split speculative- and non-speculative- sequences.
+        proposal_lens, nonzero_proposal_len_seqs, nonzero_proposal_len_indices = self._split_by_max_model_len(
+            seq_group_metadata_list, max_proposal_len)
+
+        if nonzero_proposal_len_seqs:
+            # Speculate tokens using the draft worker for the speculative
+            # sequences.
+            maybe_sampler_output = self._draft_worker.execute_model_multi_step(
+                seq_group_metadata_list=nonzero_proposal_len_seqs,
+                blocks_to_swap_in=blocks_to_swap_in,
+                blocks_to_swap_out=blocks_to_swap_out,
+                blocks_to_copy=blocks_to_copy,
+                num_steps=max_proposal_len,
+            )
+        else:
+            # If no sequences can be speculated, set sampler output to None.
+            maybe_sampler_output = None
+
+        # Combine speculative- and non-speculative sequences into the same
+        # representation.
+        proposal_tokens, proposal_probs, proposal_lens = self._merge_outputs(
+            batch_size=len(seq_group_metadata_list),
+            max_proposal_len=max_proposal_len,
+            maybe_sampler_output=maybe_sampler_output,
+            proposal_lens=proposal_lens,
+            nonzero_proposal_len_indices=nonzero_proposal_len_indices,
+        )
+
+        proposals = SpeculativeProposals(
+            proposal_token_ids=proposal_tokens,
+            proposal_probs=proposal_probs,
+            proposal_lens=proposal_lens,
+        )
+
+        return proposals
+
+    def _split_by_max_model_len(
+        self,
+        seq_group_metadata_list: List[SequenceGroupMetadata],
+        max_proposal_len: int,
+    ) -> Tuple[List[int], List[SequenceGroupMetadata], List[int]]:
+        """Determine which sequences would exceed the max model length.
+        """
+
+        proposal_lens: List[int] = []
+        nonzero_proposal_len_seqs: List[SequenceGroupMetadata] = []
+        nonzero_proposal_len_indices: List[int] = []
+        for i, seq_group_metadata in enumerate(seq_group_metadata_list):
+            seq_data = next(iter(seq_group_metadata.seq_data.values()))
+            seq_len = seq_data.get_len()
+
+            # Currently only proposal lens of 0 or the global batch proposal len
+            # are supported.
+            if seq_len + max_proposal_len < self._max_model_len:
+                proposal_lens.append(max_proposal_len)
+                nonzero_proposal_len_seqs.append(seq_group_metadata)
+                nonzero_proposal_len_indices.append(i)
+            else:
+                proposal_lens.append(0)
+
+        return proposal_lens, nonzero_proposal_len_seqs, nonzero_proposal_len_indices
+
+    def _merge_outputs(
+        self,
+        batch_size: int,
+        max_proposal_len: int,
+        maybe_sampler_output: Optional[SamplerOutput],
+        proposal_lens: List[int],
+        nonzero_proposal_len_indices: List[int],
+    ) -> Tuple[torch.Tensor, torch.tensor, torch.Tensor]:
+        """After speculations are produced, merge the speculation results with
+        the skipped sequences.
+        """
+        if maybe_sampler_output is None:
+            # If no speculative tokens, the sampler output will be None.
+            # In this case we return empty tensors.
+            proposal_tokens = torch.zeros(0,
+                                          max_proposal_len,
+                                          dtype=torch.long,
+                                          device=self._device)
+            proposal_probs = torch.zeros(0,
+                                         max_proposal_len,
+                                         self._vocab_size,
+                                         dtype=torch.float32,
+                                         device=self._device)
+            proposal_lens = torch.zeros(len(proposal_lens),
+                                        dtype=torch.long,
+                                        device=self._device)
+            return proposal_tokens, proposal_probs, proposal_lens
+
+        sampler_output = maybe_sampler_output
+
+        proposal_tokens, proposal_probs = sampler_output_to_torch(
+            sampler_output)
+
+        # Now, reformat the output GPU tensors such that each sequence has
+        # a proposal. the proposal can be empty, e.g. [-1, -1, -1]
+
+        entire_proposal_tokens = torch.full(size=(batch_size,
+                                                  *proposal_tokens.shape[1:]),
+                                            fill_value=-1,
+                                            dtype=torch.long,
+                                            device=self._device)
+        entire_proposal_tokens[nonzero_proposal_len_indices] = proposal_tokens
+        entire_proposal_probs = torch.zeros(batch_size,
+                                            *proposal_probs.shape[1:],
+                                            dtype=torch.float32,
+                                            device=self._device)
+        entire_proposal_probs[nonzero_proposal_len_indices] = proposal_probs
+
+        proposal_tokens, proposal_probs = entire_proposal_tokens, entire_proposal_probs
+
+        proposal_lens = torch.zeros(batch_size,
+                                    dtype=torch.long,
+                                    device=self._device)
+        proposal_lens[nonzero_proposal_len_indices] = max_proposal_len
+
+        return proposal_tokens, proposal_probs, proposal_lens

vllm/spec_decode/spec_decode_worker.py

@@ -0,0 +1,372 @@
+from typing import List, Tuple, Optional, Dict
+from functools import cached_property
+
+import torch
+
+from vllm.spec_decode.metrics import AsyncMetricsCollector
+from vllm.sequence import (SamplerOutput, SequenceGroupMetadata,
+                           SequenceGroupOutput, SequenceOutput)
+from vllm.worker.worker import Worker
+from vllm.spec_decode.multi_step_worker import MultiStepWorker
+from vllm.model_executor.layers.rejection_sampler import RejectionSampler
+from vllm.config import CacheConfig
+from vllm.spec_decode.util import nvtx_range, get_all_seq_ids, split_batch_by_proposal_len
+from vllm.spec_decode.interfaces import SpeculativeProposals, SpeculativeScores
+from vllm.spec_decode.batch_expansion import BatchExpansionTop1Scorer
+from vllm.spec_decode.interfaces import SpeculativeScorer
+
+
+class SpecDecodeWorker:
+    """Worker which implements speculative decoding.
+
+    Speculative decoding reduces decoding per-token latency by using a proposal
+    method, such as a small draft model, to speculate ahead of a larger LLM. The
+    probabilities of the speculative tokens are then determined by the larger
+    LLM, after which some verification routine determines which (if any) of the
+    speculative tokens are accepted by the larger LLM.
+
+    See https://github.com/vllm-project/vllm/pull/2188 and 
+    https://github.com/vllm-project/vllm/pull/3103 for more info.
+
+    The current implementation has the following limitations:
+    * Only draft-model proposal is implemented (contributions for more forms are
+        welcome!).
+    * Only top-1 proposal and scoring are implemented. Tree-attention is left as
+        future work.
+    * Only lossless rejection sampling is supported. Contributions adding lossy
+        verification routines are welcome (e.g. Medusa's typical acceptance).
+    * All sequences in a batch must have the same proposal length, or zero. This
+        can be improved by having per-sequence speculation in the future.
+    * The scoring forward pass is done without an MQA kernel, which is
+        suboptimal especially as the batch size, proposal length, and sequence
+        lengths grow. Contributions to add a MQA scoring are welcome once
+        correctness tests pass.
+        More info here https://docs.google.com/document/d/1T-JaS2T1NRfdP51qzqpyakoCXxSXTtORppiwaj5asxA/edit.
+    """
+
+    def __init__(
+        self,
+        proposer_worker: MultiStepWorker,
+        scorer_worker: Worker,
+        rejection_sampler: RejectionSampler,
+        metrics_collector: Optional[AsyncMetricsCollector] = None,
+    ):
+        """
+        Create a SpecDecodeWorker.
+
+        Args:
+            proposer_worker: A worker that can produce speculative tokens for
+                sequences.
+            scorer_worker: A worker that produces probabilities of speculative
+                tokens according to some base model. Typically a vanilla vLLM
+                Worker.
+            rejection_sampler: A Torch module used to perform modified rejection
+                sampling for speculative decoding.
+            metrics_collector: Helper class for collecting metrics; can be set
+                for testing purposes.
+        """
+        self.proposer_worker = proposer_worker
+        self.scorer_worker = scorer_worker
+        self.rejection_sampler = rejection_sampler
+
+        self._metrics = AsyncMetricsCollector(
+            rejection_sampler
+        ) if metrics_collector is None else metrics_collector
+
+        self.probs_dtype = self.rejection_sampler.probs_dtype
+        self.token_id_dtype = self.rejection_sampler.token_id_dtype
+
+        self.scorer: SpeculativeScorer = None
+
+    def init_model(self) -> None:
+        """Initialize both scorer and proposer models.
+        """
+        # The scorer worker model is initialized first in case the proposer
+        # model has a smaller TP degree than the target worker.
+        self.scorer_worker.init_model()
+        self.proposer_worker.init_model()
+
+        self._metrics.init_gpu_tensors(self.rank)
+        self.rejection_sampler.init_gpu_tensors(self.rank)
+        self.scorer = BatchExpansionTop1Scorer(
+            scorer_worker=self.scorer_worker,
+            device=self.device,
+            vocab_size=self._vocab_size)
+
+    def profile_num_available_blocks(self, block_size: int,
+                                     gpu_memory_utilization: float,
+                                     cpu_swap_space: int,
+                                     cache_dtype: str) -> Tuple[int, int]:
+        """Determine the number of cache blocks to use.
+
+        This is done by profiling the scorer model (which is typically the
+        larger of the two). Then the total memory which would be used by the
+        scorer cache is divided evenly between the proposer and scorer model KV,
+        such that the number of blocks is equal in both KV caches.
+        """
+        num_gpu_blocks, num_cpu_blocks = (
+            self.scorer_worker.profile_num_available_blocks(
+                block_size, gpu_memory_utilization, cpu_swap_space,
+                cache_dtype))
+
+        scorer_cache_block_size_bytes = self.scorer_worker.get_cache_block_size_bytes(
+            block_size, cache_dtype)
+        proposer_cache_block_size_bytes = self.proposer_worker.get_cache_block_size_bytes(
+            block_size, cache_dtype)
+
+        new_num_gpu_blocks = split_num_cache_blocks_evenly(
+            scorer_cache_block_size_bytes, proposer_cache_block_size_bytes,
+            num_gpu_blocks)
+        return new_num_gpu_blocks, num_cpu_blocks
+
+    def init_cache_engine(self, cache_config: CacheConfig):
+        """Initialize the cache engine of the scorer and proposer workers.
+        """
+        self.scorer_worker.init_cache_engine(cache_config)
+        self.proposer_worker.init_cache_engine(cache_config)
+
+    @torch.inference_mode()
+    def execute_model(
+        self,
+        seq_group_metadata_list: List[SequenceGroupMetadata],
+        blocks_to_swap_in: Optional[Dict[int, int]],
+        blocks_to_swap_out: Optional[Dict[int, int]],
+        blocks_to_copy: Optional[Dict[int, List[int]]],
+        num_spec_tokens: int,
+    ) -> List[SamplerOutput]:
+        """Perform speculative decoding on the input batch.
+        """
+
+        assert seq_group_metadata_list is not None, (
+            "speculative decoding "
+            "requires non-None seq_group_metadata_list")
+
+        # If no spec tokens, call the proposer and scorer workers normally.
+        # Used for prefill.
+        if num_spec_tokens == 0 or len(seq_group_metadata_list) == 0:
+            return self._run_no_spec(
+                seq_group_metadata_list=seq_group_metadata_list,
+                blocks_to_swap_in=blocks_to_swap_in,
+                blocks_to_swap_out=blocks_to_swap_out,
+                blocks_to_copy=blocks_to_copy,
+            )
+
+        return self._run_speculative_decoding_step(
+            seq_group_metadata_list=seq_group_metadata_list,
+            blocks_to_swap_in=blocks_to_swap_in,
+            blocks_to_swap_out=blocks_to_swap_out,
+            blocks_to_copy=blocks_to_copy,
+            k=num_spec_tokens,
+        )
+
+    @nvtx_range("spec_decode_worker._run_no_spec")
+    def _run_no_spec(
+        self,
+        seq_group_metadata_list: List[SequenceGroupMetadata],
+        blocks_to_swap_in: Optional[Dict[int, int]],
+        blocks_to_swap_out: Optional[Dict[int, int]],
+        blocks_to_copy: Optional[Dict[int, List[int]]],
+    ) -> List[SamplerOutput]:
+        """Run a prefill step, without any speculation. The input is sent to the
+        proposer and scorer model so that the KV cache is consistent between the
+        two.
+        """
+
+        self.proposer_worker.execute_model(
+            seq_group_metadata_list=seq_group_metadata_list,
+            blocks_to_swap_in=blocks_to_swap_in,
+            blocks_to_swap_out=blocks_to_swap_out,
+            blocks_to_copy=blocks_to_copy,
+            return_python_output=False)
+
+        sampler_output = self.scorer_worker.execute_model(
+            seq_group_metadata_list=seq_group_metadata_list,
+            blocks_to_swap_in=blocks_to_swap_in,
+            blocks_to_swap_out=blocks_to_swap_out,
+            blocks_to_copy=blocks_to_copy,
+        )
+
+        # Clear device tensors from sampler output. This reduces communication
+        # overhead when the engine runs in a different process than the workers.
+        sampler_output.probs = None
+        sampler_output.sampled_tokens = None
+        return [sampler_output]
+
+    @nvtx_range("spec_decode_worker._run_speculative_decoding_step")
+    def _run_speculative_decoding_step(
+        self,
+        seq_group_metadata_list: List[SequenceGroupMetadata],
+        blocks_to_swap_in: Optional[Dict[int, int]],
+        blocks_to_swap_out: Optional[Dict[int, int]],
+        blocks_to_copy: Optional[Dict[int, List[int]]],
+        k: int,
+    ) -> List[SamplerOutput]:
+        """Execute a single step of speculative decoding.
+
+        This invokes the proposer worker to get k speculative tokens for each
+        sequence, then scores each speculative token using the scoring worker.
+
+        Returns a list of SamplerOutput, each containing a single token per
+        sequence.
+        """
+
+        # Generate proposals using draft worker.
+        proposals = self.proposer_worker.get_spec_proposals(
+            seq_group_metadata_list, blocks_to_swap_in, blocks_to_swap_out,
+            blocks_to_copy, k)
+
+        proposal_scores = self.scorer.score_proposals(
+            seq_group_metadata_list,
+            blocks_to_swap_in,
+            blocks_to_swap_out,
+            blocks_to_copy,
+            k,
+            proposals,
+        )
+
+        accepted_token_ids = self._verify_tokens(seq_group_metadata_list,
+                                                 proposal_scores, proposals, k)
+
+        return self._create_output_sampler_list(seq_group_metadata_list,
+                                                accepted_token_ids, k)
+
+    @nvtx_range("spec_decode_worker._verify_tokens")
+    def _verify_tokens(
+        self,
+        seq_group_metadata_list: List[SequenceGroupMetadata],
+        proposal_scores: SpeculativeScores,
+        proposals: SpeculativeProposals,
+        max_proposal_len: int,
+    ) -> torch.Tensor:
+        """Determine which speculative tokens are accepted using the
+        probabilities of each token according to the proposer and scorer models.
+        """
+        proposal_lens_list = proposals.proposal_lens.tolist()
+
+        # vLLM currently only supports proposal lens equal to zero or the batch
+        # proposal len. This adds some complexity (splitting the batch into spec
+        # and non spec sequences) and should be removed in the future. It can be
+        # done by supporting per-sequence proposal lens.
+        _, spec_indices = split_batch_by_proposal_len(
+            seq_group_metadata_list,
+            proposal_lens_list,
+            select_proposal_len_zero=False)
+        _, non_spec_indices = split_batch_by_proposal_len(
+            seq_group_metadata_list,
+            proposal_lens_list,
+            select_proposal_len_zero=True)
+        original_indices = spec_indices + non_spec_indices
+
+        proposal_probs = proposal_scores.probs[spec_indices, :-1]
+        bonus_token_ids = proposal_scores.token_ids[spec_indices, -1:]
+        non_spec_token_ids = proposal_scores.token_ids[non_spec_indices]
+
+        accepted_token_ids = self.rejection_sampler(
+            proposal_probs,
+            bonus_token_ids,
+            proposals.proposal_probs,
+            proposals.proposal_token_ids,
+        )
+
+        # Append output tokens from non-speculative sequences to
+        # the accepted token ids tensor.
+        non_spec_token_ids = non_spec_token_ids.expand(-1, max_proposal_len +
+                                                       1).clone()
+        non_spec_token_ids[:, 1:] = -1
+        accepted_token_ids = torch.cat(
+            [accepted_token_ids, non_spec_token_ids])
+
+        # Rearrange so that results are in the order of the original seq group
+        # metadata.
+        accepted_token_ids[original_indices] = accepted_token_ids.clone()
+
+        return accepted_token_ids
+
+    def _create_output_sampler_list(
+        self,
+        seq_group_metadata_list: List[SequenceGroupMetadata],
+        accepted_token_ids: torch.Tensor,  # shape: [batch_size, k+1]
+        k: int,
+    ) -> List[SamplerOutput]:
+        """Given the accepted token ids, create a list of SamplerOutput.
+
+        The output is padded with -1 tokens such that each sequence has
+        the same number of outputs.
+        """
+        seq_ids = get_all_seq_ids(seq_group_metadata_list)
+
+        # shape: [k+1, batch_size]
+        accepted_token_ids_by_step = accepted_token_ids.transpose(0,
+                                                                  1).tolist()
+        sampler_output_list = []
+        for token_ids_by_step in accepted_token_ids_by_step:
+            if all(token_id == -1 for token_id in token_ids_by_step):
+                break
+
+            step_output_token_ids = []
+            for token_id, seq_id in zip(token_ids_by_step, seq_ids):
+                step_output_token_ids.append(
+                    SequenceGroupOutput(
+                        samples=[
+                            SequenceOutput(
+                                parent_seq_id=seq_id,
+                                output_token=token_id,
+                                # TODO Add verifier logprobs.
+                                logprobs={token_id: 0.0},
+                            )
+                        ],
+                        prompt_logprobs=None,
+                    ))
+            sampler_output_list.append(
+                SamplerOutput(outputs=step_output_token_ids))
+
+        maybe_rejsample_metrics = self._metrics.maybe_collect_rejsample_metrics(
+            k)
+        if maybe_rejsample_metrics is not None:
+            sampler_output_list[
+                0].spec_decode_worker_metrics = maybe_rejsample_metrics
+
+        return sampler_output_list
+
+    @cached_property
+    def _vocab_size(self) -> int:
+        """Get the vocab size of the model and make sure it's consistent between
+        draft and target workers.
+        """
+        vocab_sizes = [
+            worker.vocab_size
+            for worker in [self.proposer_worker, self.scorer_worker]
+        ]
+        assert all(vocab_sizes[0] == vocab_size for vocab_size in vocab_sizes)
+        return vocab_sizes[0]
+
+    @property
+    def rank(self):
+        return self.scorer_worker.rank
+
+    @property
+    def device(self):
+        return self.scorer_worker.device
+
+
+def split_num_cache_blocks_evenly(scorer_cache_block_size_bytes: int,
+                                  proposer_cache_block_size_bytes: int,
+                                  total_num_gpu_blocks: int) -> int:
+    """Given total_num_gpu_blocks, the number of GPU blocks that could be
+    allocate to the target model, this function calculates how many blocks
+    should be given to the draft and target model.
+
+    Note that usually the block size, in bytes, of each model is different,
+    as it's a function of number of KV/layer, number of heads, and hidden
+    dimension size.
+
+    Since the target and draft models allocate the same number of blocks, we
+    simply calculate the number of blocks where if allocated by both models,
+    the total memory usage from KV cache is no larger than the number of
+    blocks allocatable by the target model alone.
+    """
+    new_num_gpu_blocks = int(
+        total_num_gpu_blocks * scorer_cache_block_size_bytes /
+        (proposer_cache_block_size_bytes + scorer_cache_block_size_bytes))
+
+    return new_num_gpu_blocks

vllm/spec_decode/util.py

@@ -0,0 +1,99 @@
+import torch
+from typing import List, Tuple
+from vllm.sequence import SequenceGroupMetadata, SamplerOutput
+from contextlib import contextmanager
+from itertools import chain
+
+SeqId = int
+
+
+def get_all_seq_ids(
+        seq_group_metadata_list: List[SequenceGroupMetadata]) -> List[SeqId]:
+    """Given a list of SequenceGroupMetadata, create a list of all
+    sequence ids.
+    """
+    return list(
+        chain.from_iterable([
+            seq_group_metadata.seq_data.keys()
+            for seq_group_metadata in seq_group_metadata_list
+        ]))
+
+
+def split_batch_by_proposal_len(
+    seq_group_metadata_list: List[SequenceGroupMetadata],
+    proposal_lens: List[int], select_proposal_len_zero: bool
+) -> Tuple[List[SequenceGroupMetadata], List[int]]:
+    """Utility function that splits a batch based on whether the proposal len is
+    zero or not. We should remove this once vLLM supports per-sequence proposal
+    lens in a batch.
+    """
+
+    if select_proposal_len_zero:
+        predicate = lambda proposal_len: proposal_len == 0
+    else:
+        predicate = lambda proposal_len: proposal_len != 0
+
+    indices = [
+        i for i, (_, proposal_len
+                  ) in enumerate(zip(seq_group_metadata_list, proposal_lens))
+        if predicate(proposal_len)
+    ]
+    seq_groups = [
+        seq_group for seq_group, proposal_len in zip(
+            seq_group_metadata_list, proposal_lens) if predicate(proposal_len)
+    ]
+
+    return seq_groups, indices
+
+
+def sampler_output_to_torch(
+    sampler_output_list: List[SamplerOutput],
+) -> Tuple[torch.Tensor, torch.Tensor]:
+    """Utility function which converts a list of SamplerOutput to tensors.
+
+        Returns:
+            sampled_token_ids: torch.Tensor
+                shape: [batch_size, len(sampler_output_list)]
+
+            sampled_token_probs: torch.Tensor
+                shape: [batch_size, len(sampler_output_list), vocab_size]
+        """
+
+    # shape: [batch_size, num_sampler_output, vocab_size]
+    sampled_token_probs = torch.stack(
+        [
+            sampler_output.sampled_token_probs
+            for sampler_output in sampler_output_list
+        ],
+        dim=0,
+    ).transpose(0, 1)
+
+    # shape: [batch_size, num_sampler_output]
+    sampled_token_ids = torch.stack(
+        [
+            sampler_output.sampled_token_ids.flatten()
+            for sampler_output in sampler_output_list
+        ],
+        dim=0,
+    ).transpose(0, 1)
+
+    return sampled_token_ids, sampled_token_probs
+
+
+@contextmanager
+def nvtx_range(msg, *args, **kwargs):
+    """ 
+    Context manager / decorator that pushes an NVTX range at the beginning
+    of its scope, and pops it at the end. If extra arguments are given,
+    they are passed as arguments to msg.format().
+
+    If running with cuda graphs, you must enable nsys cuda graph profiling.
+
+    Arguments:
+        msg (string): message to associate with the range
+    """
+    torch.cuda.nvtx.range_push(msg.format(*args, **kwargs))
+    try:
+        yield
+    finally:
+        torch.cuda.nvtx.range_pop()

vllm/worker/model_runner.py

@@ -97,8 +97,6 @@ class ModelRunner:
             f"Loading model weights took {self.model_memory_usage / float(2**30):.4f} GB"
         )
 
-        vocab_size = self.model.config.vocab_size
-
         if self.lora_config:
             assert hasattr(
                 self.model, "supported_lora_modules"
@@ -111,7 +109,7 @@ class ModelRunner:
             self.lora_manager = LRUCacheWorkerLoRAManager(
                 self.scheduler_config.max_num_seqs,
                 self.scheduler_config.max_num_batched_tokens +
-                self.scheduler_config.max_paddings, vocab_size,
+                self.scheduler_config.max_paddings, self.vocab_size,
                 self.lora_config, self.device, self.model.embedding_modules,
                 self.model.embedding_padding_modules)
             self.model = self.lora_manager.create_lora_manager(self.model)
@@ -607,8 +605,7 @@ class ModelRunner:
     @torch.inference_mode()
     def profile_run(self) -> None:
         # Enable top-k sampling to reflect the accurate memory usage.
-        vocab_size = self.model_config.get_vocab_size()
+        sampling_params = SamplingParams(top_p=0.99, top_k=self.vocab_size - 1)
-        sampling_params = SamplingParams(top_p=0.99, top_k=vocab_size - 1)
         max_num_batched_tokens = self.scheduler_config.max_num_batched_tokens
         max_num_seqs = self.scheduler_config.max_num_seqs
 
@@ -774,6 +771,10 @@ class ModelRunner:
         self.graph_runners.clear()
         self.cupy_nccl_backend = None
 
+    @property
+    def vocab_size(self) -> int:
+        return self.model_config.get_vocab_size()
+
 
 class CUDAGraphRunner:
 

vllm/worker/spec_decode/multi_step_worker.py

@@ -1,178 +0,0 @@
-from typing import List, Dict
-import copy
-
-import torch
-
-from vllm.sequence import SamplerOutput, SequenceGroupMetadata
-from vllm.worker.worker import Worker
-
-
-class MultiStepWorker(Worker):
-    """The MultiStepWorker is equivalent to a Worker except that it allows
-    multiple forward passes in a single call, assuming the scheduler has
-    allocated enough space to store the additional KV. This reduces overhead
-    by invoking the scheduler less.
-
-    The MultiStepWorker does not support cache swap operations, or beam search.
-    Cache swap operations do not require large modifications. On the other hand,
-    beam search requires memory allocations during sequence forks and thus
-    requires more thought for MultiStepWorker support.
-    """
-
-    @torch.inference_mode()
-    def execute_model_multi_step(
-        self,
-        seq_group_metadata_list: List[SequenceGroupMetadata],
-        blocks_to_swap_in: Dict[int, int],
-        blocks_to_swap_out: Dict[int, int],
-        blocks_to_copy: Dict[int, List[int]],
-        num_steps: int,
-    ) -> List[SamplerOutput]:
-        """Run the model forward pass num_steps times. Returns the list of
-        sampler output, one per model forward pass.
-        """
-        self._raise_if_unsupported(seq_group_metadata_list, blocks_to_swap_in,
-                                   blocks_to_swap_out, blocks_to_copy)
-
-        # Shallow copy input data so modifications (such as appending tokens)
-        # do not cause side-effects.
-        copied_seq_group_metadata_list = self._shallow_copy_inputs(
-            seq_group_metadata_list)
-
-        # Assert enough KV space for num_steps tokens per sequence.
-        self._assert_enough_kv_space(seq_group_metadata_list, num_steps)
-
-        # Run model num_steps times.
-        model_outputs = []
-        for _ in range(num_steps):
-            model_output = super().execute_model(
-                seq_group_metadata_list=copied_seq_group_metadata_list,
-                blocks_to_swap_in=blocks_to_swap_in,
-                blocks_to_swap_out=blocks_to_swap_out,
-                blocks_to_copy=blocks_to_copy,
-            )
-
-            self._append_new_tokens(model_output,
-                                    copied_seq_group_metadata_list)
-            model_outputs.append(model_output)
-
-        return model_outputs
-
-    def _append_new_tokens(
-            self, model_output: SamplerOutput,
-            seq_group_metadata_list: SequenceGroupMetadata) -> None:
-        """Given model output from a single run, append the tokens to the
-        sequences. This is normally done outside of the worker, but it is
-        required if the worker is to perform multiple forward passes.
-        """
-        for seq_group_metadata, sequence_group_outputs in zip(
-                seq_group_metadata_list, model_output):
-            seq_group_metadata.is_prompt = False
-
-            for seq_output in sequence_group_outputs.samples:
-                # NOTE: Beam search is not supported, so we can assume that
-                # parent_seq_id == seq_id.
-                seq = seq_group_metadata.seq_data[seq_output.parent_seq_id]
-
-                token_id = seq_output.output_token
-                token_logprob = seq_output.logprobs[token_id]
-
-                seq.append_token_id(token_id, token_logprob.logprob)
-
-    def _shallow_copy_inputs(
-        self, seq_group_metadata_list: List[SequenceGroupMetadata]
-    ) -> List[SequenceGroupMetadata]:
-        """Copy input data structures to remove side-effects when input data
-        structures are shared with other modules.
-
-        The multi-step worker must be able to append tokens to sequences after
-        a forward pass. This necessitates modification of the data structures
-        used by the worker. Since these data structures are shared with other
-        parts of vLLM, like the scheduler, we must take care not to introduce
-        unexpected side-effects.
-
-        When Ray is used to orchestrate worker processes (such as when the
-        tensor-parallel degree is >1), this is not a problem because the input
-        datastructures will be serialized and created anew in the worker
-        process.
-
-        However, when Ray is not used to orchestrate the worker processes (such
-        as when the tensor-parallel degree is 1), this is a problem. We avoid
-        the problem by shallow-copying the input datastructures (specifically,
-        the parts that will change in multiple steps).
-        """
-
-        # Shallow-copy the list of SequenceGroupMetadata. This allows us to
-        # append tokens and change is_prompt without external side-effects.
-        new_seq_group_metadata_list = []
-
-        for old_seq_group_metadata in seq_group_metadata_list:
-            # We must shallow-copy seq_group_metadata as is_prompt could change.
-            seq_group_metadata = copy.copy(old_seq_group_metadata)
-            new_seq_group_metadata_list.append(seq_group_metadata)
-
-            # We must shallow-copy seq_data as we will append token ids
-            new_seq_data = {}
-            for seq_id, old_seq_data in seq_group_metadata.seq_data.items():
-                new_seq_data[seq_id] = copy.copy(old_seq_data)
-                new_seq_data[
-                    seq_id].output_token_ids = old_seq_data.output_token_ids[:]
-
-            seq_group_metadata.seq_data = new_seq_data
-
-        return new_seq_group_metadata_list
-
-    def _assert_enough_kv_space(
-            self, seq_group_metadata_list: List[SequenceGroupMetadata],
-            num_steps: int) -> None:
-        """Assert there are enough physical blocks per sequence to store the
-        current KV plus additional KV from num_steps tokens.
-        """
-        assert self.model_runner.block_size is not None
-        for seq_group_metadata in seq_group_metadata_list:
-            # Only one seq_id is guaranteed because there is no beam search.
-            seq_id = list(seq_group_metadata.seq_data.keys())[0]
-            seq = seq_group_metadata.seq_data[seq_id]
-
-            # After num_steps, the seq len will be the current seq len
-            # plus one token per step.
-            final_seq_len = seq.get_len() + num_steps
-
-            # We will have final_seq_len - 1 KV because vLLM saves KV for a
-            # token in the iteration after the token was generated.
-            required_num_kv_slots = final_seq_len - 1
-
-            # The allocated number of kv slots is the number of allocated blocks
-            # times the number of slots of block.
-            number_physical_blocks = len(
-                seq_group_metadata.block_tables[seq_id])
-            allocated_kv_slots = (number_physical_blocks *
-                                  self.model_runner.block_size)
-
-            if required_num_kv_slots > allocated_kv_slots:
-                request_id = seq_group_metadata.request_id
-                raise ValueError(
-                    "The worker attempted to run "
-                    f"{num_steps} times but found insufficient KV space for "
-                    f"{request_id=} {seq_id=}. ({allocated_kv_slots=} "
-                    f"{required_num_kv_slots=}).")
-
-    def _raise_if_unsupported(
-        self,
-        seq_group_metadata_list: List[SequenceGroupMetadata],
-        blocks_to_swap_in: Dict[int, int],
-        blocks_to_swap_out: Dict[int, int],
-        blocks_to_copy: Dict[int, List[int]],
-    ) -> None:
-        """MultiStepWorker does not yet implement support for cache swap
-        operations or beam search.
-        """
-        if any([blocks_to_swap_in, blocks_to_swap_out, blocks_to_copy]):
-            raise NotImplementedError(
-                "MultiStepWorker does not support cache operations")
-
-        if any(
-                len(seq_group_metadata.seq_data.keys()) != 1
-                for seq_group_metadata in seq_group_metadata_list):
-            raise NotImplementedError(
-                "MultiStepWorker does not support beam search.")

vllm/worker/worker.py

@@ -130,8 +130,8 @@ class Worker:
         # GPU did not change their memory usage during the profiling.
         peak_memory = self.init_gpu_memory - free_gpu_memory
 
-        cache_block_size = CacheEngine.get_cache_block_size(
+        cache_block_size = self.get_cache_block_size_bytes(
-            block_size, cache_dtype, self.model_config, self.parallel_config)
+            block_size, cache_dtype)
         num_gpu_blocks = int(
             (total_gpu_memory * gpu_memory_utilization - peak_memory) //
             cache_block_size)
@@ -232,6 +232,22 @@ class Worker:
     def list_loras(self) -> Set[int]:
         return self.model_runner.list_loras()
 
+    @property
+    def max_model_len(self) -> int:
+        return self.model_config.max_model_len
+
+    @property
+    def vocab_size(self) -> int:
+        return self.model_runner.vocab_size
+
+    def get_cache_block_size_bytes(self, block_size: int,
+                                   cache_dtype: str) -> int:
+        """Get the size of the KV cache block size in bytes.
+        """
+        return CacheEngine.get_cache_block_size(block_size, cache_dtype,
+                                                self.model_config,
+                                                self.parallel_config)
+
 
 def init_distributed_environment(
     parallel_config: ParallelConfig,