[Speculative decoding][Re-take] Enable TP>1 speculative decoding (vllm-project#4840)

Co-authored-by: Cade Daniel <[email protected]>
Co-authored-by: Cade Daniel <[email protected]>

Author: comaniac

.buildkite/test-pipeline.yaml

@@ -42,6 +42,7 @@ steps:
   - TEST_DIST_MODEL=meta-llama/Llama-2-7b-hf DISTRIBUTED_EXECUTOR_BACKEND=mp pytest -v -s distributed/test_basic_distributed_correctness.py
   - TEST_DIST_MODEL=facebook/opt-125m DISTRIBUTED_EXECUTOR_BACKEND=mp pytest -v -s distributed/test_chunked_prefill_distributed.py
   - TEST_DIST_MODEL=meta-llama/Llama-2-7b-hf DISTRIBUTED_EXECUTOR_BACKEND=mp pytest -v -s distributed/test_chunked_prefill_distributed.py
+  - pytest -v -s spec_decode/e2e/test_integration_dist.py 
 
 - label: Distributed Tests (Multiple Groups)
   working_dir: "/vllm-workspace/tests"

benchmarks/benchmark_latency.py

@@ -18,6 +18,8 @@ def main(args: argparse.Namespace):
     # NOTE(woosuk): If the request cannot be processed in a single batch,
     # the engine will automatically process the request in multiple batches.
     llm = LLM(model=args.model,
+              speculative_model=args.speculative_model,
+              num_speculative_tokens=args.num_speculative_tokens,
               tokenizer=args.tokenizer,
               quantization=args.quantization,
               tensor_parallel_size=args.tensor_parallel_size,
@@ -28,6 +30,7 @@ def main(args: argparse.Namespace):
               quantization_param_path=args.quantization_param_path,
               device=args.device,
               ray_workers_use_nsight=args.ray_workers_use_nsight,
+              use_v2_block_manager=args.use_v2_block_manager,
               enable_chunked_prefill=args.enable_chunked_prefill,
               download_dir=args.download_dir,
               block_size=args.block_size)
@@ -99,6 +102,8 @@ def run_to_completion(profile_dir: Optional[str] = None):
         description='Benchmark the latency of processing a single batch of '
         'requests till completion.')
     parser.add_argument('--model', type=str, default='facebook/opt-125m')
+    parser.add_argument('--speculative-model', type=str, default=None)
+    parser.add_argument('--num-speculative-tokens', type=int, default=None)
     parser.add_argument('--tokenizer', type=str, default=None)
     parser.add_argument('--quantization',
                         '-q',
@@ -181,6 +186,7 @@ def run_to_completion(profile_dir: Optional[str] = None):
         action='store_true',
         help='If True, the prefill requests can be chunked based on the '
         'max_num_batched_tokens')
+    parser.add_argument('--use-v2-block-manager', action='store_true')
     parser.add_argument(
         "--ray-workers-use-nsight",
         action='store_true',

tests/spec_decode/e2e/test_compatibility.py

@@ -5,56 +5,6 @@
 from .conftest import get_output_from_llm_generator
 
 
-@pytest.mark.parametrize(
-    "common_llm_kwargs",
-    [{
-        "model": "JackFram/llama-68m",
-        "speculative_model": "JackFram/llama-68m",
-        "num_speculative_tokens": 5,
-
-        # Required for spec decode.
-        "use_v2_block_manager": True
-    }])
-@pytest.mark.parametrize(
-    "per_test_common_llm_kwargs",
-    [
-        {
-            # Expect failure as spec decode not supported by
-            # Ray backend.
-            "worker_use_ray": True,
-        },
-    ])
-@pytest.mark.parametrize("test_llm_kwargs", [{}])
-@pytest.mark.parametrize("seed", [1])
-def test_spec_decode_xfail_ray(test_llm_generator):
-    """Verify that speculative decoding with Ray fails.
-    """
-    output_len = 128
-    temperature = 0.0
-
-    prompts = [
-        "Hello, my name is",
-    ]
-
-    sampling_params = SamplingParams(
-        max_tokens=output_len,
-        ignore_eos=True,
-        temperature=temperature,
-    )
-
-    try:
-        with pytest.raises(
-                AssertionError,
-                match="Speculative decoding not yet supported for "):
-            get_output_from_llm_generator(test_llm_generator, prompts,
-                                          sampling_params)
-    finally:
-        # we need to free up ray resource,
-        # so that latter test could use the gpu we allocated here
-        import ray
-        ray.shutdown()
-
-
 @pytest.mark.parametrize(
     "common_llm_kwargs",
     [{

tests/spec_decode/e2e/test_integration.py

@@ -0,0 +1,44 @@
+"""Tests which cover integration of the speculative decoding framework with
+other features, e.g. cuda graphs.
+"""
+
+import pytest
+
+from .conftest import run_greedy_equality_correctness_test
+
+
+@pytest.mark.parametrize(
+    "common_llm_kwargs",
+    [{
+        # Required for spec decode.
+        "use_v2_block_manager": True,
+
+        # Verify equality when cuda graphs allowed.
+        "enforce_eager": False,
+        "model": "JackFram/llama-68m",
+    }])
+@pytest.mark.parametrize(
+    "per_test_common_llm_kwargs",
+    [
+        {
+            # Identical models.
+            "speculative_model": "JackFram/llama-68m",
+            "num_speculative_tokens": 5,
+        },
+    ])
+@pytest.mark.parametrize("baseline_llm_kwargs", [{}])
+@pytest.mark.parametrize("test_llm_kwargs", [{}])
+@pytest.mark.parametrize("batch_size", [8])
+@pytest.mark.parametrize("output_len", [32])
+@pytest.mark.parametrize("seed", [1])
+def test_spec_decode_cuda_graph(baseline_llm_generator, test_llm_generator,
+                                batch_size, output_len):
+    """Verify spec decode equality when cuda graphs are enabled.
+    """
+    run_greedy_equality_correctness_test(
+        baseline_llm_generator,
+        test_llm_generator,
+        batch_size,
+        max_output_len=output_len,
+        force_output_len=True,
+    )

tests/spec_decode/e2e/test_integration_dist.py

@@ -0,0 +1,65 @@
+"""Tests which cover integration of the speculative decoding framework with
+tensor parallelism.
+"""
+
+import pytest
+import torch
+
+from vllm.utils import is_hip
+
+from .conftest import run_greedy_equality_correctness_test
+
+
+@pytest.mark.skipif(torch.cuda.device_count() < 2,
+                    reason="Need at least 2 GPUs to run the test.")
+@pytest.mark.parametrize(
+    "common_llm_kwargs",
+    [{
+        "model": "JackFram/llama-68m",
+
+        # Skip cuda graph recording for fast test.
+        "enforce_eager": True,
+
+        # Required for spec decode.
+        "use_v2_block_manager": True,
+        "tensor_parallel_size": 2,
+
+        # Use AsyncLLM engine, so that the engine runs in its own process.
+        # Otherwise, since vLLM does not follow true SPMD, the test runner
+        # process will have both the engine and the rank0 worker. NCCL is not
+        # cleaned up properly, and its server host thread leaks, causing the
+        # second run of the test to fail with internal NCCL error.
+        "use_async": True,
+    }])
+@pytest.mark.parametrize("per_test_common_llm_kwargs", [{}])
+@pytest.mark.parametrize("baseline_llm_kwargs", [{}])
+@pytest.mark.parametrize("test_llm_kwargs", [
+    {
+        "speculative_model": "JackFram/llama-68m",
+        "num_speculative_tokens": 3,
+    },
+    {
+        "speculative_model": "[ngram]",
+        "num_speculative_tokens": 5,
+        "ngram_prompt_lookup_max": 3,
+    },
+])
+@pytest.mark.parametrize("batch_size", [2])
+@pytest.mark.parametrize(
+    "output_len",
+    [
+        # Use smaller output len for fast test.
+        32,
+    ])
+@pytest.mark.parametrize("seed", [1])
+def test_target_model_tp_gt_1(baseline_llm_generator, test_llm_generator,
+                              batch_size: int, output_len: int):
+    """Verify greedy equality when tensor parallelism is used.
+    """
+    if is_hip():
+        pytest.skip("hip is not well-supported yet")
+    run_greedy_equality_correctness_test(baseline_llm_generator,
+                                         test_llm_generator,
+                                         batch_size,
+                                         max_output_len=output_len,
+                                         force_output_len=True)

tests/spec_decode/e2e/test_multistep_correctness.py

@@ -611,40 +611,3 @@ def test_many_k(baseline_llm_generator, test_llm_generator, batch_size: int,
                                          batch_size,
                                          max_output_len=output_len,
                                          force_output_len=True)
-
-
-@pytest.mark.parametrize(
-    "common_llm_kwargs",
-    [{
-        # Required for spec decode.
-        "use_v2_block_manager": True,
-
-        # Verify equality when cuda graphs allowed.
-        "enforce_eager": False,
-        "model": "JackFram/llama-68m",
-    }])
-@pytest.mark.parametrize(
-    "per_test_common_llm_kwargs",
-    [
-        {
-            # Identical models.
-            "speculative_model": "JackFram/llama-68m",
-            "num_speculative_tokens": 5,
-        },
-    ])
-@pytest.mark.parametrize("baseline_llm_kwargs", [{}])
-@pytest.mark.parametrize("test_llm_kwargs", [{}])
-@pytest.mark.parametrize("batch_size", [8])
-@pytest.mark.parametrize("output_len", [32])
-@pytest.mark.parametrize("seed", [1])
-def test_spec_decode_cuda_graph(baseline_llm_generator, test_llm_generator,
-                                batch_size, output_len):
-    """Verify spec decode equality when cuda graphs are enabled.
-    """
-    run_greedy_equality_correctness_test(
-        baseline_llm_generator,
-        test_llm_generator,
-        batch_size,
-        max_output_len=output_len,
-        force_output_len=True,
-    )

vllm/distributed/communication_op.py

@@ -219,16 +219,16 @@ def broadcast_tensor_dict(
      to broadcast the metadata of the dict (e.g. dict structure, tensor sizes,
      dtypes).
     """
+    # Bypass the function if we are using only 1 GPU.
+    if (not torch.distributed.is_initialized()
+            or torch.distributed.get_world_size(group=group) == 1):
+        return tensor_dict
+
     group = group or torch.distributed.group.WORLD
     metadata_group = metadata_group or get_cpu_world_group()
     ranks = torch.distributed.get_process_group_ranks(group)
     assert src in ranks, f"Invalid src rank ({src})"
 
-    # Bypass the function if we are using only 1 GPU.
-    world_size = torch.distributed.get_world_size(group=group)
-    if world_size == 1:
-        return tensor_dict
-
     rank = torch.distributed.get_rank()
     if rank == src:
         metadata_list: List[Tuple[Any, Any]] = []

vllm/executor/gpu_executor.py

@@ -15,14 +15,13 @@ class GPUExecutor(ExecutorBase):
 
     def _init_executor(self) -> None:
         """Initialize the worker and load the model.
-
-        If speculative decoding is enabled, we instead create the speculative
-        worker.
         """
-        if self.speculative_config is None:
-            self._init_non_spec_worker()
-        else:
-            self._init_spec_worker()
+        assert self.parallel_config.world_size == 1, (
+            "GPUExecutor only supports single GPU.")
+
+        self.driver_worker = self._create_worker()
+        self.driver_worker.init_device()
+        self.driver_worker.load_model()
 
     def _get_worker_kwargs(
             self,
@@ -45,66 +44,30 @@ def _get_worker_kwargs(
             distributed_init_method=distributed_init_method,
             lora_config=self.lora_config,
             vision_language_config=self.vision_language_config,
+            speculative_config=self.speculative_config,
             is_driver_worker=rank == 0,
         )
 
     def _create_worker(self,
                        local_rank: int = 0,
                        rank: int = 0,
                        distributed_init_method: Optional[str] = None):
+
+        if self.speculative_config is None:
+            worker_module_name = "vllm.worker.worker"
+            worker_class_name = "Worker"
+        else:
+            worker_module_name = "vllm.spec_decode.spec_decode_worker"
+            worker_class_name = "create_spec_worker"
+
         wrapper = WorkerWrapperBase(
-            worker_module_name="vllm.worker.worker",
-            worker_class_name="Worker",
+            worker_module_name=worker_module_name,
+            worker_class_name=worker_class_name,
         )
         wrapper.init_worker(**self._get_worker_kwargs(local_rank, rank,
                                                       distributed_init_method))
         return wrapper.worker
 
-    def _init_non_spec_worker(self):
-        assert self.parallel_config.world_size == 1, (
-            "GPUExecutor only supports single GPU.")
-
-        self.driver_worker = self._create_worker()
-        self.driver_worker.init_device()
-        self.driver_worker.load_model()
-
-    def _init_spec_worker(self):
-        """Initialize a SpecDecodeWorker, using a draft model for proposals.
-        """
-        assert self.speculative_config is not None
-
-        from vllm.spec_decode.spec_decode_worker import SpecDecodeWorker
-
-        target_worker = self._create_worker()
-
-        draft_worker_kwargs = self._get_worker_kwargs()
-        # Override draft-model specific worker args.
-        draft_worker_kwargs.update(
-            model_config=self.speculative_config.draft_model_config,
-            parallel_config=self.speculative_config.draft_parallel_config,
-            ngram_prompt_lookup_max=self.speculative_config.
-            ngram_prompt_lookup_max,
-            ngram_prompt_lookup_min=self.speculative_config.
-            ngram_prompt_lookup_min,
-            # TODO allow draft-model specific load config.
-            #load_config=self.load_config,
-        )
-
-        spec_decode_worker = SpecDecodeWorker.create_worker(
-            scorer_worker=target_worker,
-            draft_worker_kwargs=draft_worker_kwargs,
-            disable_by_batch_size=self.speculative_config.
-            speculative_disable_by_batch_size,
-        )
-
-        assert self.parallel_config.world_size == 1, (
-            "GPUExecutor only supports single GPU.")
-
-        self.driver_worker = spec_decode_worker
-
-        # Load model handled in spec decode worker.
-        self.driver_worker.init_device()
-
     def determine_num_available_blocks(self) -> Tuple[int, int]:
         """Determine the number of available KV blocks by invoking the
         underlying worker.