vllm-cpu 0.11.0.post2__cp312-cp312-manylinux_2_17_x86_64.whl

vllm/v1/worker/tpu_worker.py

@@ -0,0 +1,346 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+"""A TPU worker class."""
+
+import os
+from typing import Any, Callable, Optional, TypeVar
+
+import torch
+import torch.distributed
+import torch.nn as nn
+
+import vllm.envs as envs
+from vllm.config import VllmConfig
+from vllm.distributed import (ensure_model_parallel_initialized,
+                              init_distributed_environment)
+from vllm.distributed.kv_transfer import (ensure_kv_transfer_initialized,
+                                          has_kv_transfer_group)
+from vllm.logger import init_logger
+from vllm.lora.request import LoRARequest
+from vllm.model_executor import set_random_seed
+from vllm.platforms import current_platform
+from vllm.platforms.tpu import USE_TPU_COMMONS
+from vllm.tasks import SupportedTask
+from vllm.utils import STR_DTYPE_TO_TORCH_DTYPE, cdiv
+from vllm.v1.core.sched.output import SchedulerOutput
+from vllm.v1.kv_cache_interface import (AttentionSpec, KVCacheConfig,
+                                        KVCacheSpec)
+from vllm.v1.outputs import ModelRunnerOutput
+from vllm.v1.utils import report_usage_stats
+from vllm.v1.worker.utils import bind_kv_cache
+
+logger = init_logger(__name__)
+
+_R = TypeVar("_R")
+
+if not USE_TPU_COMMONS:
+    logger.info("tpu_commons not found, using vLLM's TPUWorker.")
+    import torch_xla.core.xla_model as xm
+    import torch_xla.debug.profiler as xp
+    import torch_xla.runtime as xr
+
+    from vllm.v1.attention.backends.pallas import TPU_HEAD_SIZE_ALIGNMENT
+    from vllm.v1.worker.tpu_model_runner import TPUModelRunner
+
+
+class TPUWorker:
+
+    def __init__(
+        self,
+        vllm_config: VllmConfig,
+        local_rank: int,
+        rank: int,
+        distributed_init_method: str,
+        is_driver_worker: bool = False,
+    ):
+        self.is_driver_worker = is_driver_worker
+        self.vllm_config = vllm_config
+        self.model_config = vllm_config.model_config
+        self.cache_config = vllm_config.cache_config
+        self.lora_config = vllm_config.lora_config
+        self.load_config = vllm_config.load_config
+        self.parallel_config = vllm_config.parallel_config
+        self.use_spmd = envs.VLLM_XLA_USE_SPMD
+        self.original_parallel_config = None
+        if self.use_spmd:
+            # Under SPMD mode, distributed env is initialized as if there is
+            # only one worker/device.
+            self.original_parallel_config = self.parallel_config
+            self.parallel_config.tensor_parallel_size = 1
+            self.parallel_config.pipeline_parallel_size = 1
+            self.parallel_config.world_size = 1
+        self.scheduler_config = vllm_config.scheduler_config
+        self.device_config = vllm_config.device_config
+        self.speculative_config = vllm_config.speculative_config
+        self.observability_config = vllm_config.observability_config
+
+        self.parallel_config.rank = rank
+        self.local_rank = local_rank
+        self.rank = rank
+        self.distributed_init_method = distributed_init_method
+
+        if self.cache_config.cache_dtype == "auto":
+            self.cache_dtype = self.model_config.dtype
+        else:
+            self.cache_dtype = STR_DTYPE_TO_TORCH_DTYPE[
+                self.cache_config.cache_dtype]
+
+        if self.model_config.trust_remote_code:
+            # note: lazy import to avoid importing torch before initializing
+            from vllm.utils import init_cached_hf_modules
+            init_cached_hf_modules()
+
+        # Delay profiler initialization to the start of the profiling.
+        # This is because in vLLM V1, MP runtime is initialized before the
+        # TPU Worker is initialized. The profiler server needs to start after
+        # MP runtime is initialized.
+        self.profiler = None
+        self.profile_dir = None
+        if envs.VLLM_TORCH_PROFILER_DIR and self.rank < 1:
+            # For TPU, we can only have 1 active profiler session for 1 profiler
+            # server. So we only profile on rank0.
+            self.profile_dir = envs.VLLM_TORCH_PROFILER_DIR
+            logger.info("Profiling enabled. Traces will be saved to: %s",
+                        self.profile_dir)
+
+        if self.model_config.seed is None:
+            self.model_config.seed = 0
+
+    def initialize_cache(self, num_gpu_blocks: int,
+                         num_cpu_blocks: int) -> None:
+        self.cache_config.num_gpu_blocks = num_gpu_blocks
+        self.cache_config.num_cpu_blocks = num_cpu_blocks
+
+    def init_device(self):
+        os.environ["PJRT_DEVICE"] = "TPU"
+        # Note: Currently the XLA compiler wrongly uses 2D ring strategy on 1D
+        # ring, the xla tpu compiler flag
+        # `xla_tpu_force_1d_allreduce_at_chunk_count` is a temporary solution to
+        # fix this. It will be removed after the bug in XLA compiler is fixed.
+        os.environ["LIBTPU_INIT_ARGS"] = (
+            os.environ.get("LIBTPU_INIT_ARGS", "") +
+            " --xla_tpu_force_1d_allreduce_at_chunk_count=1"
+            " --xla_jf_conv_input_fusion=False")
+        # --xla_jf_conv_input_fusion=False is used to improve the perf of
+        # quantized matmul.
+        torch.set_grad_enabled(False)
+        torch.set_default_dtype(self.model_config.dtype)
+
+        # Initialize the distributed environment.
+        self._init_tpu_worker_distributed_environment(
+            self.vllm_config, self.rank, self.distributed_init_method,
+            self.local_rank)
+
+        # Device initialization should happen after initializing
+        # the distributed runtime.
+        self.device = xm.xla_device()
+        self.device_config.device = self.device
+
+        # Set random seed.
+        set_random_seed(self.model_config.seed)
+        if self.model_config.seed is not None:
+            xm.set_rng_state(self.model_config.seed, self.device)
+
+        # Increase the cache size limit, which is the maximum number of
+        # dynamo graphs that can be compiled.
+        # TODO (NickLucche) On gsm we compile 80+ graphs.
+        # Re-evaluate limit, with MM we may get close to this limit.
+        torch._dynamo.config.cache_size_limit = 128
+        # Use persistent cache to avoid XLA recompilation.
+        # NOTE(woosuk): Set per-rank cache path since different ranks
+        # can have slightly different XLA graphs.
+        world_size = self.parallel_config.world_size
+        rank = xr.global_ordinal()
+        # The PyTorch/XLA compilation cache uses the Torch IR to generate keys.
+        # Consequently, changes in optimization flags, which affect compilation
+        # results, don't change the cache key. This can result in the wrong
+        # compilation being used. To prevent this, disabling the XLA compilation
+        # cache during development is recommended.We can disable it by
+        # `export VLLM_XLA_CACHE_PATH=`
+        if envs.VLLM_XLA_CACHE_PATH:
+            per_rank_path = os.path.join(envs.VLLM_XLA_CACHE_PATH,
+                                         f"tp{world_size}_rank{rank}")
+            xr.initialize_cache(per_rank_path, readonly=False)
+
+        # Init ModelRunner here, so that we have access to self.device.
+        self.model_runner = \
+            TPUModelRunner(self.vllm_config, self.device,
+                           self.original_parallel_config)
+
+        if rank == 0:
+            # If usage stat is enabled, collect relevant info.
+            report_usage_stats(self.vllm_config)
+
+    def determine_available_memory(self) -> int:
+        kv_caches: dict[str, torch.Tensor] = {}
+        kv_cache_spec = self.model_runner.get_kv_cache_spec()
+        for layer_name, layer_spec in kv_cache_spec.items():
+            if isinstance(layer_spec, AttentionSpec):
+                dtype = layer_spec.dtype
+
+                # Use an empty tensor instead of `None`` to force Dynamo to pass
+                # it by reference, rather by specializing on the value ``None``.
+                tpu_kv_cache = torch.tensor([], dtype=dtype).to(self.device)
+                kv_caches[layer_name] = tpu_kv_cache
+            else:
+                raise NotImplementedError(
+                    f"Unsupported KV cache spec '{type(layer_spec)}'")
+
+        runner_kv_caches: list[torch.Tensor] = []
+        bind_kv_cache(
+            kv_caches,
+            self.vllm_config.compilation_config.static_forward_context,
+            runner_kv_caches)
+
+        # `max_num_tokens >= max_num_batched_tokens` due to padding.
+        with self.model_runner.maybe_setup_dummy_loras(self.lora_config):
+            self.model_runner.profile_run(self.model_runner.max_num_tokens)
+
+        # Synchronize before measuring the memory usage.
+        xm.wait_device_ops()
+
+        # During the profiling run, the model runs without KV cache. After
+        # the profiling run, the model always runs with KV cache. Here we clear
+        # the dynamo cache and cached bytecode to ensure the model always has
+        # one compiled bytecode. Having one FX graph/cached bytecode per
+        # compiled model is required for `support_torch_compile` decorator to
+        # skip dynamo guard.
+        self.model_runner.reset_dynamo_cache()
+
+        # Get the maximum amount of memory used by the model weights and
+        # intermediate activations.
+        if self.use_spmd:
+            # This is a workaround for the TPU SPMD mode. The get_memory_info
+            # API doesn't work with SPMD mode in PyTorch/XLA.
+            # TODO: use xm.get_memory_info for SPMD once it's supported in
+            # PyTorch/XLA.
+            import tpu_info
+            chip_type, _ = tpu_info.device.get_local_chips()
+            device_usage = tpu_info.metrics.get_chip_usage(chip_type)
+            total_memory_size = device_usage[0].total_memory
+            current_mem = device_usage[0].memory_usage
+        else:
+            m = xm.get_memory_info(self.device)
+            total_memory_size = m["bytes_limit"]
+            current_mem = m["bytes_used"]
+        # Ideally we would use profiled = m["peak_bytes_used"] to
+        # get weights + activations. But there is memory used during
+        # compilation / weight loading that impacts the peak and
+        # there is no way to reset peak memory in XLA, So we
+        # use the heuristic of 2% of weights.
+        profiled = current_mem * 1.02
+
+        # Calculate the TPU KV cache size based on profiling.
+        usable_memory_size = int(total_memory_size *
+                                 self.cache_config.gpu_memory_utilization)
+        tpu_kv_cache_bytes = max(usable_memory_size - profiled, 0)
+        head_size = self.model_config.get_head_size()
+        if head_size > 0:
+            padded_head_size = cdiv(
+                head_size, TPU_HEAD_SIZE_ALIGNMENT) * TPU_HEAD_SIZE_ALIGNMENT
+            if padded_head_size != head_size:
+                logger.warning_once("head size is padded to %d",
+                                    padded_head_size)
+            # We adjust the usable memory size for the KV cache to prevent OOM
+            # errors, even after padding the head_size.
+            tpu_kv_cache_bytes = (tpu_kv_cache_bytes * head_size //
+                                  padded_head_size)
+        return int(tpu_kv_cache_bytes)
+
+    def execute_model(
+        self,
+        scheduler_output: "SchedulerOutput",
+    ) -> Optional[ModelRunnerOutput]:
+        output = self.model_runner.execute_model(scheduler_output)
+        # every worker's output is needed when kv_transfer_group is set up
+        return output if self.is_driver_worker or has_kv_transfer_group(
+        ) else None
+
+    def profile(self, is_start: bool = True):
+        if self.rank < 1:
+            if self.profile_dir is None:
+                raise RuntimeError("Profiler is not enabled.")
+            if is_start:
+                if self.profiler is None:
+                    self.profiler = xp.start_server(9012)
+                xp.start_trace(self.profile_dir)
+            else:
+                xp.stop_trace()
+
+    def add_lora(self, lora_request: LoRARequest) -> bool:
+        return self.model_runner.add_lora(lora_request)
+
+    def load_model(self) -> None:
+        self.model_runner.load_model()
+
+    def update_config(self, overrides: dict[str, Any]) -> None:
+        self.model_runner.update_config(overrides)
+
+    def reload_weights(self) -> None:
+        self.model_runner.reload_weights()
+
+    def compile_or_warm_up_model(self) -> None:
+        if not self.model_config.enforce_eager:
+            self.model_runner.capture_model()
+
+        # Reset the seed to ensure that the random state is not affected by
+        # the model initialization and profiling.
+        set_random_seed(self.model_config.seed)
+
+    def get_model(self) -> nn.Module:
+        return self.model_runner.get_model()
+
+    def get_supported_tasks(self) -> tuple[SupportedTask, ...]:
+        return self.model_runner.get_supported_tasks()
+
+    def get_kv_cache_spec(self) -> dict[str, KVCacheSpec]:
+        return self.model_runner.get_kv_cache_spec()
+
+    def initialize_from_config(self, kv_cache_config: KVCacheConfig) -> None:
+        """Allocate GPU KV cache with the specified kv_cache_config."""
+        self.model_runner.initialize_kv_cache(kv_cache_config)
+
+    def check_health(self) -> None:
+        # worker will always be healthy as long as it's running.
+        return
+
+    def _init_tpu_worker_distributed_environment(
+        self,
+        vllm_config: VllmConfig,
+        rank: int,
+        distributed_init_method: Optional[str] = None,
+        local_rank: int = -1,
+    ) -> None:
+        """Initialize the distributed environment."""
+        if self.use_spmd:
+            xr.use_spmd()
+        # NOTE(woosuk): This is just to initialize the TP group and broadcast
+        # the input objects on CPU. The all-reduce and all-gather ops on TPU
+        # are invoked by `xm.all_reduce` and `xm.all_gather` which use their
+        # own context.
+        parallel_config = vllm_config.parallel_config
+        init_distributed_environment(
+            world_size=parallel_config.world_size,
+            rank=rank,
+            local_rank=local_rank,
+            distributed_init_method=distributed_init_method,
+            backend=current_platform.dist_backend,
+        )
+        ensure_model_parallel_initialized(
+            parallel_config.tensor_parallel_size,
+            parallel_config.pipeline_parallel_size)
+
+        ensure_kv_transfer_initialized(vllm_config)
+
+    def shutdown(self) -> None:
+        self.model_runner.ensure_kv_transfer_shutdown()
+
+    def apply_model(self, fn: Callable[[nn.Module], _R]) -> _R:
+        """Apply a function on the model inside this worker."""
+        return fn(self.get_model())
+
+
+if USE_TPU_COMMONS:
+    from tpu_commons.worker import TPUWorker as TPUCommonsWorker
+
+    TPUWorker = TPUCommonsWorker  # type: ignore

vllm/v1/worker/ubatch_splitting.py

@@ -0,0 +1,192 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+from typing import Optional
+
+import numpy as np
+import torch
+
+from vllm.config import ParallelConfig, VllmConfig
+from vllm.forward_context import DPMetadata
+from vllm.logger import init_logger
+from vllm.utils import round_up
+from vllm.v1.worker.ubatch_utils import (UBatchSlice, UBatchSlices,
+                                         is_second_ubatch_empty)
+
+logger = init_logger(__name__)
+
+
+def should_ubatch_with_num_tokens(
+    should_ubatch: bool,
+    orig_num_tokens_per_ubatch: int,
+    padded_num_tokens_per_ubatch: int,
+    vllm_config: VllmConfig,
+) -> tuple[bool, Optional[torch.Tensor]]:
+    dp_size = vllm_config.parallel_config.data_parallel_size
+    dp_rank = vllm_config.parallel_config.data_parallel_rank
+    return DPMetadata.should_ubatch_across_dp(should_ubatch,
+                                              orig_num_tokens_per_ubatch,
+                                              padded_num_tokens_per_ubatch,
+                                              dp_size, dp_rank)
+
+
+def check_ubatch_thresholds(config: ParallelConfig, num_tokens: int,
+                            uniform_decode: bool) -> bool:
+    if not config.enable_dbo:
+        return False
+    if uniform_decode:
+        return num_tokens >= config.dbo_decode_token_threshold
+    else:
+        return num_tokens >= config.dbo_prefill_token_threshold
+
+
+def get_dp_padding_ubatch(
+        num_tokens_unpadded: int, num_tokens_padded: int,
+        should_attempt_ubatching: bool,
+        vllm_config: VllmConfig) -> tuple[bool, Optional[torch.Tensor]]:
+    """
+    1. Decides if each DP rank is going to microbatch. Either all ranks
+    run with microbatching or none of them do. If this function decides
+    not to run with microbatching. It will "abort" meaning that no padding
+    information will be returned to the caller. It will return (False, None)
+
+    2. Determines the total number of tokens that each rank will run.
+    All ranks will be padded out so that the run with the same number
+    of tokens
+
+    Returns: tuple[
+        should_ubatch: Are all DP ranks going to microbatch
+        num_tokens_after_padding: A tensor containing the total number of
+        tokens per-microbatch for each DP rank including padding. Will be
+        None if should_ubatch if False
+    ]
+
+    """
+    assert num_tokens_padded >= num_tokens_unpadded
+    dp_size = vllm_config.parallel_config.data_parallel_size
+    if dp_size == 1:
+        # Early exit.
+        return False, None
+
+    # If this DP rank doesn't want to attempt microbatching
+    if not should_attempt_ubatching:
+        (should_ubatch, num_tokens_across_dp) = should_ubatch_with_num_tokens(
+            False, 0, 0, vllm_config)
+        assert should_ubatch is False
+        assert num_tokens_across_dp is None
+        return should_ubatch, num_tokens_across_dp
+
+    # Round up to the next multiple of two for even divisibility
+    num_tokens_padded = round_up(num_tokens_padded, 2)
+    num_tokens_per_ubatch = num_tokens_padded // 2
+    should_ubatch = True
+
+    # Sanity Check that the existing padding isn't giving us an empty second
+    # ubatch. Abort if so
+    if is_second_ubatch_empty(num_tokens_unpadded, num_tokens_padded):
+        logger.debug(
+            "Empty second µbatch detected: unpadded tokens: %s, padded "
+            "tokens: %s", num_tokens_unpadded, num_tokens_padded)
+        should_ubatch = False
+
+    # Note that we compute the number of padded tokens per ubatch
+    (should_ubatch, num_tokens_across_dp) = should_ubatch_with_num_tokens(
+        should_ubatch, num_tokens_unpadded // 2, num_tokens_per_ubatch,
+        vllm_config)
+    if not should_ubatch:
+        assert num_tokens_across_dp is None
+        return should_ubatch, num_tokens_across_dp
+
+    assert num_tokens_across_dp is not None
+
+    max_tokens_across_dp_cpu = int(torch.max(num_tokens_across_dp).item())
+    num_tokens_after_padding = torch.tensor([max_tokens_across_dp_cpu] *
+                                            dp_size,
+                                            device="cpu",
+                                            dtype=torch.int32)
+    return should_ubatch, num_tokens_after_padding
+
+def create_ubatch_slices(num_scheduled_tokens: np.ndarray, split_point: int) \
+    -> UBatchSlices:
+    # TODO(lucas): Refactor the gpu_model_runner.py so we can pass
+    # in cu_num_tokens directly (i.e. query_start_loc)
+    cu_num_tokens = np.zeros(len(num_scheduled_tokens) + 1, dtype=np.int32)
+    np.cumsum(num_scheduled_tokens, dtype=np.int32, out=cu_num_tokens[1:])
+
+    first_ubatch_token_slice = slice(0, split_point)
+    second_ubatch_token_slice = slice(split_point, cu_num_tokens[-1])
+
+    # Determine request slices using exclusive stop semantics
+    # First ubatch includes requests whose tokens overlap [0, split_point)
+    first_ubatch_req_stop = int(
+        np.searchsorted(cu_num_tokens, split_point, side="left"))
+    first_ubatch_req_slice = slice(0, first_ubatch_req_stop)
+
+    # Second ubatch starts at the request that contains the split_point
+    # or the request starting exactly at split_point (if on boundary)
+    second_ubatch_req_start = int(
+        np.searchsorted(cu_num_tokens, split_point, side="right") - 1)
+    second_ubatch_req_slice = slice(second_ubatch_req_start,
+                                    len(cu_num_tokens) - 1)
+
+    return [
+        UBatchSlice(first_ubatch_req_slice, first_ubatch_token_slice),
+        UBatchSlice(second_ubatch_req_slice, second_ubatch_token_slice)
+    ]
+
+
+def ubatch_split(
+    num_scheduled_tokens_per_request: np.ndarray,
+    num_tokens_unpadded: int,
+    num_tokens_padded: int,
+    uniform_decode: bool,
+    vllm_config: VllmConfig,
+) -> tuple[Optional[UBatchSlices], Optional[torch.Tensor]]:
+    """
+    Coordinates amongst all DP ranks to determine if and how the full batch
+    should be split into microbatches.
+
+    Returns: tuple[
+        ubatch_slices: if this is set then all DP ranks have agreed to 
+        microbatch
+        num_tokens_after_padding: A tensor containing the total number of
+        tokens per-microbatch for each DP rank including padding. Will be
+        None if ubatch_slices is None
+    ]
+
+    """
+    parallel_config = vllm_config.parallel_config
+    # Don't bother with the should_ubatch handshaking unless microbatching
+    # is enabled
+    if not parallel_config.enable_dbo:
+        return (None, None)
+
+    # Check preconditions for microbatching
+    should_attempt_ubatching = check_ubatch_thresholds(
+        parallel_config,
+        num_tokens_unpadded,
+        uniform_decode=uniform_decode,
+    )
+
+    # Don't microbatch unless every other DP worker is also microbatching
+    should_ubatch, num_tokens_after_padding = get_dp_padding_ubatch(
+        num_tokens_unpadded,
+        num_tokens_padded,
+        should_attempt_ubatching,
+        vllm_config,
+    )
+
+    if not should_ubatch:
+        return (None, None)
+
+    # This doesn't actually pad the ubatch slices. It just initializes the
+    # split point to the padded value so that padding can be applied
+    # to the second ubatch in pad_out_ubatch_slice after attention
+    # metadata creation
+    assert num_tokens_after_padding is not None
+    token_split_point = int(num_tokens_after_padding[0].item())
+
+    ubatch_slices = create_ubatch_slices(num_scheduled_tokens_per_request,
+                                         token_split_point)
+
+    return (ubatch_slices, num_tokens_after_padding)

vllm/v1/worker/ubatch_utils.py

@@ -0,0 +1,27 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+from dataclasses import dataclass
+
+from typing_extensions import TypeAlias
+
+
+@dataclass
+class UBatchSlice:
+    request_slice: slice
+    token_slice: slice
+
+    def is_empty(self) -> bool:
+        return self.request_slice.start == self.request_slice.stop \
+            or self.token_slice.start == self.token_slice.stop
+
+    @property
+    def num_tokens(self) -> int:
+        return self.token_slice.stop - self.token_slice.start
+
+
+UBatchSlices: TypeAlias = list[UBatchSlice]
+
+
+def is_second_ubatch_empty(orig_num_tokens_per_ubatch: int,
+                           padded_num_tokens_per_ubatch: int) -> bool:
+    return padded_num_tokens_per_ubatch >= 2 * orig_num_tokens_per_ubatch