104 lines
3.4 KiB
Python
104 lines
3.4 KiB
Python
# SPDX-License-Identifier: Apache-2.0
|
|
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
|
"""Sequence and its related classes."""
|
|
|
|
from dataclasses import dataclass
|
|
from typing import TYPE_CHECKING, Any
|
|
|
|
import msgspec
|
|
import torch
|
|
|
|
if TYPE_CHECKING:
|
|
from vllm.v1.worker.kv_connector_model_runner_mixin import KVConnectorOutput
|
|
else:
|
|
KVConnectorOutput = Any
|
|
|
|
VLLM_TOKEN_ID_ARRAY_TYPE = "l"
|
|
|
|
VLLM_INVALID_TOKEN_ID = -1
|
|
|
|
|
|
@dataclass
|
|
class RequestMetrics:
|
|
"""Metrics associated with a request.
|
|
|
|
Attributes:
|
|
arrival_time: The time when the request arrived.
|
|
first_scheduled_time: The time when the request was first scheduled.
|
|
first_token_time: The time when the first token was generated.
|
|
time_in_queue: The time the request spent in the queue.
|
|
finished_time: The time when the request was finished.
|
|
scheduler_time: The time spent in the scheduler when this request was
|
|
being considered by the scheduler.
|
|
model_forward_time: The time spent in the model forward pass when this
|
|
request was in the batch.
|
|
model_execute_time: The time spent in the model execute function. This
|
|
will include model forward, block/sync across
|
|
workers, cpu-gpu sync time and sampling time.
|
|
"""
|
|
|
|
arrival_time: float
|
|
last_token_time: float
|
|
first_scheduled_time: float | None
|
|
first_token_time: float | None
|
|
time_in_queue: float | None
|
|
finished_time: float | None = None
|
|
scheduler_time: float | None = None
|
|
model_forward_time: float | None = None
|
|
model_execute_time: float | None = None
|
|
|
|
|
|
# cannot use msgspec.Struct here because Dynamo does not support it
|
|
@dataclass
|
|
class IntermediateTensors:
|
|
"""For all pipeline stages except the last, we need to return the hidden
|
|
states and residuals to be sent to the next stage. This data structure
|
|
contains the hidden states and residuals for a request.
|
|
|
|
Each stage also needs to handle its own kv_connector_output.
|
|
"""
|
|
|
|
tensors: dict[str, torch.Tensor]
|
|
kv_connector_output: KVConnectorOutput | None
|
|
|
|
def __init__(self, tensors):
|
|
# manually define this function, so that
|
|
# Dynamo knows `IntermediateTensors()` comes from this file.
|
|
# Otherwise, dataclass will generate this function by evaluating
|
|
# a string, and we will lose the information about the source file.
|
|
self.tensors = tensors
|
|
|
|
def __getitem__(self, key: str | slice):
|
|
if isinstance(key, str):
|
|
return self.tensors[key]
|
|
elif isinstance(key, slice):
|
|
return self.__class__({k: v[key] for k, v in self.tensors.items()})
|
|
|
|
def __setitem__(self, key: str, value: torch.Tensor):
|
|
self.tensors[key] = value
|
|
|
|
def items(self):
|
|
return self.tensors.items()
|
|
|
|
def __len__(self):
|
|
return len(self.tensors)
|
|
|
|
def __eq__(self, other: object):
|
|
if not isinstance(other, self.__class__):
|
|
return False
|
|
if self.tensors.keys() != other.tensors.keys():
|
|
return False
|
|
return all(torch.equal(self.tensors[k], other.tensors[k]) for k in self.tensors)
|
|
|
|
def __repr__(self) -> str:
|
|
return f"IntermediateTensors(tensors={self.tensors})"
|
|
|
|
|
|
class ExecuteModelRequest(
|
|
msgspec.Struct,
|
|
array_like=True, # type: ignore[call-arg]
|
|
omit_defaults=True,
|
|
): # type: ignore[call-arg]
|
|
# Placeholder. Remove.
|
|
pass
|