Make tests/lora/utils usable by plugins

tests/lora/test_layers.py

@@ -30,6 +30,7 @@
 )
 from vllm.lora.models import LoRALayerWeights, PackedLoRALayerWeights
 from vllm.lora.punica_wrapper import get_punica_wrapper
+from vllm.lora.utils import DummyLoRAManager
 from vllm.model_executor.layers.linear import (
     ColumnParallelLinear,
     MergedColumnParallelLinear,
@@ -46,8 +47,6 @@
 from vllm.model_executor.utils import set_random_seed
 from vllm.platforms import current_platform
 
-from .utils import DummyLoRAManager
-
 TOLERANCES = {
     torch.float16: (5e-3, 5e-3),
     torch.float32: (5e-3, 5e-3),

tests/lora/test_lora_manager.py

@@ -24,11 +24,10 @@
 )
 from vllm.lora.peft_helper import PEFTHelper
 from vllm.lora.request import LoRARequest
+from vllm.lora.utils import create_peft_lora
 from vllm.lora.worker_manager import LRUCacheWorkerLoRAManager, WorkerLoRAManager
 from vllm.platforms import current_platform
 
-from .utils import create_peft_lora
-
 EMBEDDING_MODULES = {
     "embed_tokens": "input_embeddings",
     "lm_head": "output_embeddings",

tests/lora/test_punica_ops.py

@@ -9,10 +9,9 @@
 import vllm.lora.ops.triton_ops as triton_ops
 from vllm.lora.ops.triton_ops import LoRAKernelMeta
 from vllm.lora.ops.triton_ops.utils import _LORA_A_PTR_DICT, _LORA_B_PTR_DICT
+from vllm.lora.utils import PunicaTensors, assert_close, generate_data_for_nslices
 from vllm.platforms import current_platform
 
-from .utils import PunicaTensors, assert_close, generate_data_for_nslices
-
 
 @pytest.fixture(autouse=True)
 def reset_device(reset_default_device):