Move scheduler code from tp_worker.py to scheduler.py (#1538)

This commit is contained in:
Lianmin Zheng
2024-09-29 17:42:45 -07:00
committed by GitHub
parent acaffd233f
commit f86c1e611f
8 changed files with 933 additions and 870 deletions

View File

@@ -18,7 +18,6 @@ The definition of objects transfered between different
processes (TokenizerManager, DetokenizerManager, Controller).
"""
import copy
import uuid
from dataclasses import dataclass
from typing import Dict, List, Optional, Union
@@ -53,12 +52,12 @@ class GenerateReqInput:
stream: bool = False
# The modalities of the image data [image, multi-images, video]
modalities: Optional[List[str]] = None
is_single: bool = True
# LoRA related
lora_path: Optional[Union[List[Optional[str]], Optional[str]]] = None
# Whether it is a single request or a batch request
is_single: bool = True
def post_init(self):
if (self.text is None and self.input_ids is None) or (
self.text is not None and self.input_ids is not None
@@ -307,10 +306,6 @@ class BatchTokenIDOut:
meta_info: List[Dict]
finished_reason: List[BaseFinishReason]
def __post_init__(self):
# deepcopy meta_info to avoid modification in place
self.meta_info = copy.deepcopy(self.meta_info)
@dataclass
class BatchStrOut: