biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Misc] Enable V1 LoRA by default (#15320)

Browse Source 
Signed-off-by: Varun Sundar Rabindranath <varun@neuralmagic.com>
Co-authored-by: Varun Sundar Rabindranath <varun@neuralmagic.com>
This commit is contained in:
Varun Sundar Rabindranath
2025-04-01 04:53:56 -04:00
committed by GitHub
parent 30d6a015e0
commit 79455cf421
12 changed files with 125 additions and 87 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
tests/lora/test_lora_manager.py
View File

@@ -7,7 +7,6 @@ import torch
from safetensors.torch import load_file
from torch import nn
from vllm import envs
from vllm.config import LoRAConfig
from vllm.lora.layers import (ColumnParallelLinearWithLoRA,
MergedColumnParallelLinearWithLoRA,
@@ -33,6 +32,17 @@ DEVICES = ([
] if current_platform.is_cuda_alike() else ["cpu"])
@pytest.fixture(scope="function", autouse=True)
def use_v0_only(monkeypatch: pytest.MonkeyPatch):
"""
Some tests depend on V0 internals. Since both V0 and V1 use the same
LoRAModelManager it is okay to just test V0.
"""
with monkeypatch.context() as m:
m.setenv('VLLM_USE_V1', '0')
yield
@pytest.mark.parametrize("device", DEVICES)
def test_from_lora_tensors(sql_lora_files, device):
tensors = load_file(
@@ -411,7 +421,6 @@ def test_lru_lora_model_manager(dist_init, dummy_model, device):
assert manager.device == device
@pytest.mark.skipif(envs.VLLM_USE_V1, reason="Test leverages V0 internals.")
@pytest.mark.parametrize("device", DEVICES)
def test_lru_cache_worker_adapter_manager(llama_2_7b_model_extra_embeddings,
sql_lora_files, device):
@@ -491,7 +500,6 @@ def test_lru_cache_worker_adapter_manager(llama_2_7b_model_extra_embeddings,
device)
@pytest.mark.skipif(envs.VLLM_USE_V1, reason="Test leverages V0 internals.")
@pytest.mark.parametrize("device", DEVICES)
def test_worker_adapter_manager(llama_2_7b_model_extra_embeddings,
sql_lora_files, device):