Files
vllm-omni/vllm_omni/core/sched/output.py
T

76 lines
2.6 KiB
Python
Raw Normal View History

from dataclasses import dataclass, field
2025-10-22 15:05:23 +08:00
from vllm.v1.core.sched.output import CachedRequestData, NewRequestData, SchedulerOutput
2025-10-24 16:49:19 +08:00
from vllm.v1.request import Request
2025-11-18 20:54:26 +08:00
2025-10-24 16:49:19 +08:00
from vllm_omni.engine import AdditionalInformationPayload, PromptEmbedsPayload
2025-10-22 15:05:23 +08:00
@dataclass
class OmniNewRequestData(NewRequestData):
"""New request data for omni models with embeddings support.
Extends NewRequestData to include prompt embeddings and additional
information for direct transfer between pipeline stages.
Args:
prompt_embeds: Optional serialized prompt embeddings payload
additional_information: Optional serialized additional information
dictionary containing tensors or lists
"""
2025-10-22 15:05:23 +08:00
# Optional serialized prompt embeddings
prompt_embeds: PromptEmbedsPayload | None = None
# Optional external request ID for tracking
external_req_id: str | None = None
2025-10-22 15:05:23 +08:00
# Optional serialized additional information
additional_information: AdditionalInformationPayload | None = None
2025-10-22 15:05:23 +08:00
@classmethod
def from_request(
cls,
request: Request,
block_ids: tuple[list[int], ...],
2025-12-19 16:42:02 +08:00
prefill_token_ids: list[int] | None = None,
2025-10-24 10:41:54 +08:00
) -> "OmniNewRequestData":
"""Create OmniNewRequestData from a Request object.
Args:
request: Request object to convert
block_ids: Tuple of block ID lists for KV cache allocation
Returns:
OmniNewRequestData instance with data from the request
"""
2025-10-22 15:05:23 +08:00
return cls(
req_id=request.request_id,
external_req_id=request.external_req_id,
2025-10-22 15:05:23 +08:00
prompt_token_ids=request.prompt_token_ids,
2025-11-04 22:06:41 +08:00
mm_features=request.mm_features,
2025-10-22 15:05:23 +08:00
sampling_params=request.sampling_params,
pooling_params=request.pooling_params,
block_ids=block_ids,
num_computed_tokens=request.num_computed_tokens,
lora_request=request.lora_request,
prompt_embeds=request.prompt_embeds,
2025-12-19 16:42:02 +08:00
prefill_token_ids=prefill_token_ids,
2025-10-22 15:05:23 +08:00
additional_information=request.additional_information,
2025-10-24 16:49:19 +08:00
)
@dataclass
class OmniCachedRequestData(CachedRequestData):
"""Cached request data for omni models with embeddings support.
Args:
prompt_token_ids: Mapping from request ID to list of prompt token IDs
"""
prompt_token_ids: dict[str, list[int]]
@dataclass
class OmniSchedulerOutput(SchedulerOutput):
"""Scheduler output with omni-specific transfer metadata."""
finished_requests_needing_kv_transfer: dict[str, dict] = field(default_factory=dict)