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

[LoRA] Cleanup LoRA unused code (#29611)

Browse Source 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>
This commit is contained in:
Jee Jee Li
2025-11-29 14:52:58 +08:00
committed by GitHub
parent 4a80ad0a25
commit 39e63dec7c
46 changed files with 126 additions and 173 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
tests/entrypoints/conftest.py
View File

@@ -188,11 +188,11 @@ number: "1" | "2"
@pytest.fixture(scope="session")
def zephyr_lora_files():
"""Download zephyr LoRA files once per test session."""
def qwen3_lora_files():
"""Download Qwen3 LoRA files once per test session."""
from huggingface_hub import snapshot_download
return snapshot_download(repo_id="typeof/zephyr-7b-beta-lora")
return snapshot_download(repo_id="charent/self_cognition_Alice")
@pytest.fixture(scope="session")

2
tests/entrypoints/openai/test_basic.py
View File

@@ -16,7 +16,7 @@ from vllm.version import __version__ as VLLM_VERSION
from ...utils import RemoteOpenAIServer
MODEL_NAME = "HuggingFaceH4/zephyr-7b-beta"
MODEL_NAME = "Qwen/Qwen3-0.6B"
@pytest.fixture(scope="module")

8
tests/entrypoints/openai/test_chat.py
View File

@@ -19,6 +19,14 @@ from ...utils import RemoteOpenAIServer
MODEL_NAME = "HuggingFaceH4/zephyr-7b-beta"
@pytest.fixture(scope="module")
def zephyr_lora_files():
"""Download zephyr LoRA files once per test session."""
from huggingface_hub import snapshot_download
return snapshot_download(repo_id="typeof/zephyr-7b-beta-lora")
@pytest.fixture(scope="module")
def server(zephyr_lora_files): # noqa: F811
args = [

3
tests/entrypoints/openai/test_chunked_prompt.py
View File

@@ -8,7 +8,7 @@ import pytest_asyncio
from ...utils import RemoteOpenAIServer
# any model with a chat template should work here
MODEL_NAME = "HuggingFaceH4/zephyr-7b-beta"
MODEL_NAME = "Qwen/Qwen3-0.6B"
@pytest.fixture(scope="module")
@@ -20,7 +20,6 @@ def server():
"--max-model-len",
"8192",
"--enforce-eager",
# lora config below
"--max-num-seqs",
"128",
"--enable-chunked-prefill",

43
tests/entrypoints/openai/test_lora_adapters.py
View File

@@ -13,9 +13,8 @@ import pytest_asyncio
from ...utils import RemoteOpenAIServer
# any model with a chat template should work here
MODEL_NAME = "HuggingFaceH4/zephyr-7b-beta"
# technically this needs Mistral-7B-v0.1 as base, but we're not testing
# generation quality here
MODEL_NAME = "Qwen/Qwen3-0.6B"
BADREQUEST_CASES = [
(
@@ -33,11 +32,11 @@ BADREQUEST_CASES = [
@pytest.fixture(scope="module", params=[True])
def server_with_lora_modules_json(request, zephyr_lora_files):
def server_with_lora_modules_json(request, qwen3_lora_files):
# Define the json format LoRA module configurations
lora_module_1 = {
"name": "zephyr-lora",
"path": zephyr_lora_files,
"name": "qwen3-lora",
"path": qwen3_lora_files,
"base_model_name": MODEL_NAME,
}
@@ -74,7 +73,7 @@ async def client(server_with_lora_modules_json):
@pytest.mark.asyncio
async def test_static_lora_lineage(client: openai.AsyncOpenAI, zephyr_lora_files):
async def test_static_lora_lineage(client: openai.AsyncOpenAI, qwen3_lora_files):
models = await client.models.list()
models = models.data
served_model = models[0]
@@ -82,17 +81,17 @@ async def test_static_lora_lineage(client: openai.AsyncOpenAI, zephyr_lora_files
assert served_model.id == MODEL_NAME
assert served_model.root == MODEL_NAME
assert served_model.parent is None
assert all(lora_model.root == zephyr_lora_files for lora_model in lora_models)
assert all(lora_model.root == qwen3_lora_files for lora_model in lora_models)
assert all(lora_model.parent == MODEL_NAME for lora_model in lora_models)
assert lora_models[0].id == "zephyr-lora"
assert lora_models[0].id == "qwen3-lora"
@pytest.mark.asyncio
async def test_dynamic_lora_lineage(client: openai.AsyncOpenAI, zephyr_lora_files):
async def test_dynamic_lora_lineage(client: openai.AsyncOpenAI, qwen3_lora_files):
response = await client.post(
"load_lora_adapter",
cast_to=str,
body={"lora_name": "zephyr-lora-3", "lora_path": zephyr_lora_files},
body={"lora_name": "qwen3-lora-3", "lora_path": qwen3_lora_files},
)
# Ensure adapter loads before querying /models
assert "success" in response
@@ -100,9 +99,9 @@ async def test_dynamic_lora_lineage(client: openai.AsyncOpenAI, zephyr_lora_file
models = await client.models.list()
models = models.data
dynamic_lora_model = models[-1]
assert dynamic_lora_model.root == zephyr_lora_files
assert dynamic_lora_model.root == qwen3_lora_files
assert dynamic_lora_model.parent == MODEL_NAME
assert dynamic_lora_model.id == "zephyr-lora-3"
assert dynamic_lora_model.id == "qwen3-lora-3"
@pytest.mark.asyncio
@@ -134,7 +133,7 @@ async def test_dynamic_lora_invalid_files(client: openai.AsyncOpenAI, tmp_path):
async def test_dynamic_lora_badrequests(
client: openai.AsyncOpenAI,
tmp_path,
zephyr_lora_files,
qwen3_lora_files,
test_name: str,
config_change: dict,
expected_error: str,
@@ -143,7 +142,7 @@ async def test_dynamic_lora_badrequests(
test_dir = tmp_path / test_name
# Copy adapter files
shutil.copytree(zephyr_lora_files, test_dir)
shutil.copytree(qwen3_lora_files, test_dir)
# Load and modify configuration
config_path = test_dir / "adapter_config.json"
@@ -167,7 +166,7 @@ async def test_dynamic_lora_badrequests(
@pytest.mark.asyncio
async def test_multiple_lora_adapters(
client: openai.AsyncOpenAI, tmp_path, zephyr_lora_files
client: openai.AsyncOpenAI, tmp_path, qwen3_lora_files
):
"""Validate that many loras can be dynamically registered and inferenced
with concurrently"""
@@ -178,7 +177,7 @@ async def test_multiple_lora_adapters(
await client.post(
"load_lora_adapter",
cast_to=str,
body={"lora_name": adapter_name, "lora_path": str(zephyr_lora_files)},
body={"lora_name": adapter_name, "lora_path": str(qwen3_lora_files)},
)
for _ in range(3):
await client.completions.create(
@@ -199,7 +198,7 @@ async def test_multiple_lora_adapters(
@pytest.mark.asyncio
async def test_loading_invalid_adapters_does_not_break_others(
client: openai.AsyncOpenAI, tmp_path, zephyr_lora_files
client: openai.AsyncOpenAI, tmp_path, qwen3_lora_files
):
invalid_files = tmp_path / "invalid_files"
invalid_files.mkdir()
@@ -215,7 +214,7 @@ async def test_loading_invalid_adapters_does_not_break_others(
while not stop_good_requests_event.is_set():
try:
batch = await client.completions.create(
model="zephyr-lora",
model="qwen3-lora",
prompt=["Hello there", "Foo bar bazz buzz"],
max_tokens=5,
)
@@ -254,7 +253,7 @@ async def test_loading_invalid_adapters_does_not_break_others(
await client.post(
"load_lora_adapter",
cast_to=str,
body={"lora_name": "valid", "lora_path": zephyr_lora_files},
body={"lora_name": "valid", "lora_path": qwen3_lora_files},
)
await client.completions.create(
model="valid",
@@ -267,7 +266,7 @@ async def test_loading_invalid_adapters_does_not_break_others(
async def test_beam_search_with_lora_adapters(
client: openai.AsyncOpenAI,
tmp_path,
zephyr_lora_files,
qwen3_lora_files,
):
"""Validate that async beam search can be used with lora."""
@@ -275,7 +274,7 @@ async def test_beam_search_with_lora_adapters(
await client.post(
"load_lora_adapter",
cast_to=str,
body={"lora_name": adapter_name, "lora_path": str(zephyr_lora_files)},
body={"lora_name": adapter_name, "lora_path": str(qwen3_lora_files)},
)
for _ in range(3):
await client.completions.create(

12
tests/entrypoints/openai/test_models.py
View File

@@ -8,13 +8,13 @@ import pytest_asyncio
from ...utils import RemoteOpenAIServer
# any model with a chat template should work here
MODEL_NAME = "HuggingFaceH4/zephyr-7b-beta"
MODEL_NAME = "Qwen/Qwen3-0.6B"
# technically this needs Mistral-7B-v0.1 as base, but we're not testing
# generation quality here
@pytest.fixture(scope="module")
def server(zephyr_lora_files):
def server(qwen3_lora_files):
args = [
# use half precision for speed and memory savings in CI environment
"--dtype",
@@ -25,7 +25,7 @@ def server(zephyr_lora_files):
# lora config below
"--enable-lora",
"--lora-modules",
f"zephyr-lora={zephyr_lora_files}",
f"qwen3-lora={qwen3_lora_files}",
"--max-lora-rank",
"64",
"--max-cpu-loras",
@@ -45,12 +45,12 @@ async def client(server):
@pytest.mark.asyncio
async def test_check_models(client: openai.AsyncOpenAI, zephyr_lora_files):
async def test_check_models(client: openai.AsyncOpenAI, qwen3_lora_files):
models = await client.models.list()
models = models.data
served_model = models[0]
lora_models = models[1:]
assert served_model.id == MODEL_NAME
assert served_model.root == MODEL_NAME
assert all(lora_model.root == zephyr_lora_files for lora_model in lora_models)
assert lora_models[0].id == "zephyr-lora"
assert all(lora_model.root == qwen3_lora_files for lora_model in lora_models)
assert lora_models[0].id == "qwen3-lora"

5
tests/entrypoints/openai/test_orca_metrics.py
View File

@@ -8,7 +8,7 @@ import pytest_asyncio
from ...utils import RemoteOpenAIServer
# any model with a chat template should work here
MODEL_NAME = "HuggingFaceH4/zephyr-7b-beta"
MODEL_NAME = "Qwen/Qwen3-0.6B"
@pytest.fixture(scope="module")
@@ -110,8 +110,9 @@ async def test_single_completion(client: openai.AsyncOpenAI):
choice = completion.choices[0]
assert len(choice.text) >= 5
assert choice.finish_reason == "length"
# When using Qwen3-0.6B, prompt tokens=[9707, 11, 847, 829, 374]
assert completion.usage == openai.types.CompletionUsage(
completion_tokens=5, prompt_tokens=6, total_tokens=11
completion_tokens=5, prompt_tokens=5, total_tokens=10
)
# test using token IDs

6
tests/entrypoints/openai/test_return_tokens_as_ids.py
View File

@@ -11,11 +11,11 @@ from vllm.transformers_utils.tokenizer import get_tokenizer
from ...utils import RemoteOpenAIServer
MODEL_NAME = "HuggingFaceH4/zephyr-7b-beta"
MODEL_NAME = "Qwen/Qwen3-0.6B"
@pytest.fixture(scope="module")
def default_server_args(zephyr_lora_files):
def default_server_args(qwen3_lora_files):
return [
# use half precision for speed and memory savings in CI environment
"--dtype",
@@ -28,7 +28,7 @@ def default_server_args(zephyr_lora_files):
# lora config
"--enable-lora",
"--lora-modules",
f"zephyr-lora={zephyr_lora_files}",
f"qwen3-lora={qwen3_lora_files}",
"--max-lora-rank",
"64",
"--max-cpu-loras",

2
tests/entrypoints/openai/test_tokenization.py
View File

@@ -10,7 +10,7 @@ from vllm.transformers_utils.tokenizer import get_tokenizer
from ...utils import RemoteOpenAIServer
# any model with a chat template should work here
MODEL_NAME = "HuggingFaceH4/zephyr-7b-beta"
MODEL_NAME = "Qwen/Qwen3-0.6B"
@pytest.fixture(scope="module")

2
tests/entrypoints/openai/test_uds.py
View File

@@ -10,7 +10,7 @@ from vllm.version import __version__ as VLLM_VERSION
from ...utils import RemoteOpenAIServer
MODEL_NAME = "HuggingFaceH4/zephyr-7b-beta"
MODEL_NAME = "Qwen/Qwen3-0.6B"
@pytest.fixture(scope="module")

1
tests/entrypoints/sagemaker/conftest.py
View File

@@ -9,7 +9,6 @@ import pytest_asyncio
from ...utils import RemoteOpenAIServer
# Model name constants used across tests
MODEL_NAME_ZEPHYR = "HuggingFaceH4/zephyr-7b-beta"
MODEL_NAME_SMOLLM = "HuggingFaceTB/SmolLM2-135M-Instruct"
LORA_ADAPTER_NAME_SMOLLM = "jekunz/smollm-135m-lora-fineweb-faroese"

27
tests/lora/conftest.py
View File

@@ -154,23 +154,6 @@ def dummy_model_gate_up() -> nn.Module:
return model
@pytest.fixture(scope="session")
def llama_2_7b_base_huggingface_id():
# used as a base model for testing with sql lora adapter
return "meta-llama/Llama-2-7b-hf"
@pytest.fixture(scope="session")
def sql_lora_huggingface_id():
# huggingface repo id is used to test lora runtime downloading.
return "yard1/llama-2-7b-sql-lora-test"
@pytest.fixture(scope="session")
def sql_lora_files(sql_lora_huggingface_id):
return snapshot_download(repo_id=sql_lora_huggingface_id)
@pytest.fixture(scope="session")
def mixtral_lora_files():
# Note: this module has incorrect adapter_config.json to test
@@ -256,8 +239,14 @@ def qwen3_lora_files():
@pytest.fixture(scope="session")
def llama32_lora_files():
return snapshot_download(repo_id="jeeejeee/llama32-3b-text2sql-spider")
def llama32_lora_huggingface_id():
# huggingface repo id is used to test lora runtime downloading.
return "jeeejeee/llama32-3b-text2sql-spider"
@pytest.fixture(scope="session")
def llama32_lora_files(llama32_lora_huggingface_id):
return snapshot_download(repo_id=llama32_lora_huggingface_id)
@pytest.fixture

21
tests/lora/test_lora_checkpoints.py
View File

@@ -26,8 +26,7 @@ def test_load_checkpoints(
chatglm3_lora_files,
):
packed_modules_mapping = BaiChuanBaseForCausalLM.packed_modules_mapping
embedding_modules = BaiChuanBaseForCausalLM.embedding_modules
embed_padding_modules = BaiChuanBaseForCausalLM.embedding_padding_modules
expected_lora_lst: list[str] = []
for module in BAICHUAN_LORA_MODULES:
if module in packed_modules_mapping:
@@ -47,8 +46,7 @@ def test_load_checkpoints(
peft_helper=peft_helper,
lora_model_id=1,
device="cpu",
embedding_modules=embedding_modules,
embedding_padding_modules=embed_padding_modules,
model_vocab_size=64000,
)
elif lora_name == "baichuan7B-zero":
# Test that the target_modules contain prefix
@@ -63,8 +61,7 @@ def test_load_checkpoints(
peft_helper=peft_helper,
lora_model_id=1,
device="cpu",
embedding_modules=embedding_modules,
embedding_padding_modules=embed_padding_modules,
model_vocab_size=64000,
)
elif lora_name == "baichuan7B-zero-regex":
# Test that the `target_modules` in the form of regular expressions,
@@ -78,8 +75,7 @@ def test_load_checkpoints(
peft_helper=peft_helper,
lora_model_id=1,
device="cpu",
embedding_modules=embedding_modules,
embedding_padding_modules=embed_padding_modules,
model_vocab_size=64000,
)
else:
# For the baichuan7B model, load chatglm3-6b's LoRA,
@@ -95,15 +91,13 @@ def test_load_checkpoints(
peft_helper=peft_helper,
lora_model_id=1,
device="cpu",
embedding_modules=embedding_modules,
embedding_padding_modules=embed_padding_modules,
model_vocab_size=64000,
)
def test_lora_weights_mapping(baichuan_lora_files):
packed_modules_mapping = BaiChuanBaseForCausalLM.packed_modules_mapping
embedding_modules = BaiChuanBaseForCausalLM.embedding_modules
embed_padding_modules = BaiChuanBaseForCausalLM.embedding_padding_modules
expected_lora_lst: list[str] = []
for module in BAICHUAN_LORA_MODULES:
if module in packed_modules_mapping:
@@ -128,8 +122,7 @@ def test_lora_weights_mapping(baichuan_lora_files):
peft_helper=peft_helper,
lora_model_id=1,
device="cpu",
embedding_modules=embedding_modules,
embedding_padding_modules=embed_padding_modules,
model_vocab_size=64000,
weights_mapper=hf_to_vllm_mapper,
)
for name in lora_model.loras:

11
tests/lora/test_lora_huggingface.py
View File

@@ -6,10 +6,10 @@ import pytest
from vllm.lora.models import LoRAModel
from vllm.lora.peft_helper import PEFTHelper
from vllm.lora.utils import get_adapter_absolute_path
from vllm.model_executor.models.llama import LlamaForCausalLM
from vllm.model_executor.models.qwen3 import Qwen3ForCausalLM
# Provide absolute path and huggingface lora ids
lora_fixture_name = ["sql_lora_files", "sql_lora_huggingface_id"]
lora_fixture_name = ["llama32_lora_files", "llama32_lora_huggingface_id"]
LLAMA_LORA_MODULES = [
"qkv_proj",
"o_proj",
@@ -23,9 +23,8 @@ LLAMA_LORA_MODULES = [
@pytest.mark.parametrize("lora_fixture_name", lora_fixture_name)
def test_load_checkpoints_from_huggingface(lora_fixture_name, request):
lora_name = request.getfixturevalue(lora_fixture_name)
packed_modules_mapping = LlamaForCausalLM.packed_modules_mapping
embedding_modules = LlamaForCausalLM.embedding_modules
embed_padding_modules = LlamaForCausalLM.embedding_padding_modules
packed_modules_mapping = Qwen3ForCausalLM.packed_modules_mapping
expected_lora_lst: list[str] = []
for module in LLAMA_LORA_MODULES:
if module in packed_modules_mapping:
@@ -43,8 +42,6 @@ def test_load_checkpoints_from_huggingface(lora_fixture_name, request):
peft_helper=peft_helper,
lora_model_id=1,
device="cpu",
embedding_modules=embedding_modules,
embedding_padding_modules=embed_padding_modules,
)
# Assertions to ensure the model is loaded correctly

17
tests/lora/test_lora_manager.py
View File

@@ -34,7 +34,6 @@ EMBEDDING_MODULES = {
"lm_head": "output_embeddings",
}
EMBEDDING_PADDING_MODULES = ["lm_head"]
DEVICES = (
[f"cuda:{i}" for i in range(1 if torch.cuda.device_count() == 1 else 2)]
@@ -46,24 +45,22 @@ DEFAULT_DTYPE = torch.get_default_dtype()
@pytest.mark.parametrize("device", DEVICES)
def test_from_lora_tensors(sql_lora_files, device):
tensors = load_file(os.path.join(sql_lora_files, "adapter_model.safetensors"))
def test_from_lora_tensors(qwen3_lora_files, device):
tensors = load_file(os.path.join(qwen3_lora_files, "adapter_model.safetensors"))
peft_helper = PEFTHelper.from_local_dir(
sql_lora_files, max_position_embeddings=4096
qwen3_lora_files, max_position_embeddings=4096
)
lora_model = LoRAModel.from_lora_tensors(
1,
tensors,
peft_helper=peft_helper,
device=device,
embedding_modules=EMBEDDING_MODULES,
embedding_padding_modules=EMBEDDING_PADDING_MODULES,
)
for module_name, lora in lora_model.loras.items():
assert lora.module_name == module_name
assert lora.rank == 8
assert lora.lora_alpha == 16
assert lora.lora_alpha == 32
assert lora.lora_a is not None
assert lora.lora_b is not None
assert lora.lora_a.device == torch.device(device)
@@ -430,7 +427,7 @@ def test_lru_cache_worker_adapter_manager(dist_init, dummy_model, device, tmp_pa
vllm_config.scheduler_config.max_num_seqs = 4
vllm_config.scheduler_config.max_num_batched_tokens = 2
worker_adapter_manager = LRUCacheWorkerLoRAManager(
vllm_config, device, EMBEDDING_MODULES, EMBEDDING_PADDING_MODULES
vllm_config, device, EMBEDDING_MODULES
)
worker_adapter_manager.max_num_seqs = 4
@@ -533,9 +530,7 @@ def test_worker_adapter_manager(dist_init, dummy_model_gate_up, device, tmp_path
vllm_config.scheduler_config.max_num_seqs = 4
vllm_config.scheduler_config.max_num_batched_tokens = 2
worker_adapter_manager = WorkerLoRAManager(
vllm_config, device, EMBEDDING_MODULES, EMBEDDING_PADDING_MODULES
)
worker_adapter_manager = WorkerLoRAManager(vllm_config, device, EMBEDDING_MODULES)
worker_adapter_manager.vocab_size = dummy_model_gate_up.unpadded_vocab_size
worker_adapter_manager.create_lora_manager(dummy_model_gate_up)

22
tests/lora/test_olmoe_tp.py
View File

@@ -40,7 +40,10 @@ EXPECTED_BASE_MODEL_OUTPUT = [
def generate_and_test(
llm: vllm.LLM, lora_path: str, lora_id: list[int | None] | int | None
llm: vllm.LLM,
lora_path: str,
lora_id: list[int | None] | int | None,
compare_lower: bool = False,
) -> None:
prompts = [
PROMPT_TEMPLATE.format(context="How many candidates are there?"),
@@ -74,12 +77,18 @@ def generate_and_test(
for i in range(len(EXPECTED_LORA_OUTPUT)):
req_lora_id = lora_id[i] if isinstance(lora_id, list) else lora_id
generated_text = generated_texts[i]
expected_output = (
EXPECTED_LORA_OUTPUT[i]
if req_lora_id is not None
else EXPECTED_BASE_MODEL_OUTPUT[i]
)
assert generated_texts[i].startswith(expected_output)
if compare_lower:
generated_text = generated_text.lower()
expected_output = expected_output.lower()
assert generated_text.startswith(expected_output)
def test_olmoe_lora(olmoe_lora_files):
@@ -146,6 +155,9 @@ def test_olmoe_lora_tp4(olmoe_lora_files, fully_sharded_loras):
tensor_parallel_size=4,
fully_sharded_loras=fully_sharded_loras,
)
generate_and_test(llm, olmoe_lora_files, lora_id=1)
generate_and_test(llm, olmoe_lora_files, lora_id=2)
generate_and_test(
llm, olmoe_lora_files, lora_id=1, compare_lower=fully_sharded_loras
)
generate_and_test(
llm, olmoe_lora_files, lora_id=2, compare_lower=fully_sharded_loras
)

28
tests/lora/test_peft_helper.py
View File

@@ -25,31 +25,33 @@ ERROR_CASES = [
]
def test_peft_helper_pass(sql_lora_files, tmp_path):
def test_peft_helper_pass(llama32_lora_files, tmp_path):
peft_helper = PEFTHelper.from_local_dir(
sql_lora_files, max_position_embeddings=4096
llama32_lora_files, max_position_embeddings=4096
)
lora_config = LoRAConfig(max_lora_rank=16, max_cpu_loras=3, max_loras=2)
peft_helper.validate_legal(lora_config)
assert peft_helper.r == 8
assert peft_helper.lora_alpha == 16
assert peft_helper.target_modules == [
"q_proj",
"v_proj",
"k_proj",
"o_proj",
"gate_proj",
"up_proj",
assert peft_helper.lora_alpha == 32
target_modules = sorted(peft_helper.target_modules)
assert target_modules == [
"down_proj",
"embed_tokens",
"gate_proj",
"k_proj",
"lm_head",
"o_proj",
"q_proj",
"up_proj",
"v_proj",
]
assert peft_helper.vllm_max_position_embeddings == 4096
# test RSLoRA
rslora_config = dict(use_rslora=True)
test_dir = tmp_path / "test_rslora"
shutil.copytree(sql_lora_files, test_dir)
shutil.copytree(llama32_lora_files, test_dir)
# Load and modify configuration
config_path = test_dir / "adapter_config.json"
@@ -70,14 +72,14 @@ def test_peft_helper_pass(sql_lora_files, tmp_path):
@pytest.mark.parametrize("test_name,config_change,expected_error", ERROR_CASES)
def test_peft_helper_error(
sql_lora_files,
llama32_lora_files,
tmp_path,
test_name: str,
config_change: dict,
expected_error: str,
):
test_dir = tmp_path / test_name
shutil.copytree(sql_lora_files, test_dir)
shutil.copytree(llama32_lora_files, test_dir)
# Load and modify configuration
config_path = test_dir / "adapter_config.json"

10
tests/plugins/lora_resolvers/test_filesystem_resolver.py
View File

@@ -8,8 +8,8 @@ from huggingface_hub import snapshot_download
from vllm.plugins.lora_resolvers.filesystem_resolver import FilesystemResolver
MODEL_NAME = "mistralai/Mistral-7B-v0.1"
LORA_NAME = "typeof/zephyr-7b-beta-lora"
MODEL_NAME = "Qwen/Qwen3-0.6B"
LORA_NAME = "charent/self_cognition_Alice"
PA_NAME = "swapnilbp/llama_tweet_ptune"
@@ -21,7 +21,7 @@ def adapter_cache(request, tmpdir_factory):
@pytest.fixture(scope="module")
def zephyr_lora_files():
def qwen3_lora_files():
return snapshot_download(repo_id=LORA_NAME)
@@ -31,9 +31,9 @@ def pa_files():
@pytest.mark.asyncio
async def test_filesystem_resolver(adapter_cache, zephyr_lora_files):
async def test_filesystem_resolver(adapter_cache, qwen3_lora_files):
model_files = adapter_cache / LORA_NAME
shutil.copytree(zephyr_lora_files, model_files)
shutil.copytree(qwen3_lora_files, model_files)
fs_resolver = FilesystemResolver(adapter_cache)
assert fs_resolver is not None