A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 16, 22:00 UTC
[Core] Optimize `SequenceStatus.is_finished` by switching to IntEnum (#5974)
Antoni Baum committed
Jun 29, 2024 at 05:47 UTC
7c01f706418d593b3cf23d2ec9110dca7151c539
1 file changed
+11
-14
vllm/sequence.py
+11
-14
@@ -39,24 +39,21 @@ PromptLogprobs = List[Optional[Dict[int, Logprob]]]
39
SampleLogprobs = List[Dict[int, Logprob]]
40
41
42
-class SequenceStatus(enum.Enum):
42
+class SequenceStatus(enum.IntEnum):
43
"""Status of a sequence."""
44
- WAITING = enum.auto()
45
- RUNNING = enum.auto()
46
- SWAPPED = enum.auto()
47
- FINISHED_STOPPED = enum.auto()
48
- FINISHED_LENGTH_CAPPED = enum.auto()
49
- FINISHED_ABORTED = enum.auto()
50
- FINISHED_IGNORED = enum.auto()
44
+ WAITING = 0
45
+ RUNNING = 1
46
+ SWAPPED = 2
47
+ # Note: anything after SWAPPED (2) will be considered
48
+ # as a finished status.
49
+ FINISHED_STOPPED = 3
50
+ FINISHED_LENGTH_CAPPED = 4
51
+ FINISHED_ABORTED = 5
52
+ FINISHED_IGNORED = 6
53
54
@staticmethod
55
def is_finished(status: "SequenceStatus") -> bool:
54
- return status in [
55
- SequenceStatus.FINISHED_STOPPED,
56
- SequenceStatus.FINISHED_LENGTH_CAPPED,
57
- SequenceStatus.FINISHED_ABORTED,
58
- SequenceStatus.FINISHED_IGNORED,
59
- ]
56
+ return status > SequenceStatus.SWAPPED
57
58
@staticmethod
59
def get_finished_reason(status: "SequenceStatus") -> Union[str, None]: