youngkingdom/vllm
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[Core][Distributed] code deduplication in tp&pp with coordinator(#5293)

Browse Source 
[Core][Distributed] add coordinator to reduce code duplication in tp and pp (#5293)
This commit is contained in:
youkaichao
2024-06-12 17:27:08 -07:00
committed by GitHub
parent 2135cacb45
commit ea3890a5f0
12 changed files with 625 additions and 585 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/worker/test_model_runner.py
View File

@ -1,7 +1,8 @@
import pytest
import torch
from vllm.distributed.parallel_state import init_distributed_environment
from vllm.distributed.parallel_state import (ensure_model_parallel_initialized,
init_distributed_environment)
from vllm.engine.arg_utils import EngineArgs
from vllm.model_executor.sampling_metadata import SamplingMetadata
from vllm.sequence import SamplingParams, SequenceData, SequenceGroupMetadata
@ -292,6 +293,7 @@ def distributed_init():
rank=0,
distributed_init_method=f"tcp://127.0.0.1:{get_open_port()}",
local_rank=0)
ensure_model_parallel_initialized(1, 1)
@pytest.mark.parametrize("batch_size", list(range(2, 128)))