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

[V0 Deprecation] Remove V0 LoRA test (#23418)

Browse Source 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
This commit is contained in:
Jee Jee Li
2025-08-22 17:56:51 +08:00
committed by GitHub
parent 88016c372a
commit 285178b3b8
7 changed files with 158 additions and 116 deletions
Download Patch File Download Diff File Expand all files Collapse all files

130
tests/lora/test_lora_manager.py
View File

@@ -21,6 +21,8 @@ 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",
@@ -35,17 +37,6 @@ DEVICES = ([
DEFAULT_DTYPE = torch.get_default_dtype()
@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(
@@ -326,7 +317,6 @@ def test_lru_lora_model_manager(dist_init, dummy_model, device):
max_loras=2,
lora_dtype=DEFAULT_DTYPE),
device=device)
assert all(x is None for x in manager.lora_index_to_id)
# Add up to capacity
@@ -430,32 +420,40 @@ def test_lru_lora_model_manager(dist_init, dummy_model, device):
@pytest.mark.parametrize("device", DEVICES)
def test_lru_cache_worker_adapter_manager(llama_2_7b_model_extra_embeddings,
sql_lora_files, device):
def test_lru_cache_worker_adapter_manager(dist_init, dummy_model, device,
tmp_path):
lora_config = LoRAConfig(max_lora_rank=8,
max_cpu_loras=4,
max_loras=4,
lora_dtype=DEFAULT_DTYPE)
dummy_lora_files = f"{tmp_path}/lora_adapter"
os.makedirs(dummy_lora_files, exist_ok=True)
create_peft_lora(
dummy_model,
save_dir=dummy_lora_files,
target_modules=["layer1.dense1", "dense2"],
lora_dtype=DEFAULT_DTYPE,
)
worker_adapter_manager = LRUCacheWorkerLoRAManager(
4, 2, llama_2_7b_model_extra_embeddings.unpadded_vocab_size -
lora_config.lora_extra_vocab_size, lora_config, device,
EMBEDDING_MODULES, EMBEDDING_PADDING_MODULES)
worker_adapter_manager.create_lora_manager(
llama_2_7b_model_extra_embeddings)
4, 2,
dummy_model.unpadded_vocab_size - lora_config.lora_extra_vocab_size,
lora_config, device, EMBEDDING_MODULES, EMBEDDING_PADDING_MODULES)
worker_adapter_manager.create_lora_manager(dummy_model)
mapping = LoRAMapping([], [])
worker_adapter_manager.set_active_adapters([
LoRARequest("1", 1, sql_lora_files),
LoRARequest("2", 2, sql_lora_files)
LoRARequest("1", 1, dummy_lora_files),
LoRARequest("2", 2, dummy_lora_files)
], mapping)
assert worker_adapter_manager.list_adapters() == {1, 2}
assert worker_adapter_manager._adapter_manager.lora_index_to_id[0] == 1
assert worker_adapter_manager._adapter_manager.lora_index_to_id[1] == 2
worker_adapter_manager.set_active_adapters([
LoRARequest("1", 1, sql_lora_files),
LoRARequest("3", 3, sql_lora_files),
LoRARequest("4", 4, sql_lora_files)
LoRARequest("1", 1, dummy_lora_files),
LoRARequest("3", 3, dummy_lora_files),
LoRARequest("4", 4, dummy_lora_files)
], mapping)
assert worker_adapter_manager.list_adapters() == {1, 2, 3, 4}
assert worker_adapter_manager._adapter_manager.lora_index_to_id[0] == 1
@@ -464,9 +462,9 @@ def test_lru_cache_worker_adapter_manager(llama_2_7b_model_extra_embeddings,
assert worker_adapter_manager._adapter_manager.lora_index_to_id[3] == 4
worker_adapter_manager.set_active_adapters([
LoRARequest("1", 1, sql_lora_files),
LoRARequest("2", 2, sql_lora_files),
LoRARequest("5", 5, sql_lora_files)
LoRARequest("1", 1, dummy_lora_files),
LoRARequest("2", 2, dummy_lora_files),
LoRARequest("5", 5, dummy_lora_files)
], mapping)
assert worker_adapter_manager.list_adapters() == {1, 2, 4, 5}
assert worker_adapter_manager._adapter_manager.lora_index_to_id[0] == 1
@@ -475,9 +473,9 @@ def test_lru_cache_worker_adapter_manager(llama_2_7b_model_extra_embeddings,
assert worker_adapter_manager._adapter_manager.lora_index_to_id[3] == 4
worker_adapter_manager.set_active_adapters([
LoRARequest("1", 1, sql_lora_files),
LoRARequest("1", 1, sql_lora_files),
LoRARequest("1", 1, sql_lora_files)
LoRARequest("1", 1, dummy_lora_files),
LoRARequest("1", 1, dummy_lora_files),
LoRARequest("1", 1, dummy_lora_files)
], mapping)
assert worker_adapter_manager.list_adapters() == {1, 2, 4, 5}
assert worker_adapter_manager._adapter_manager.lora_index_to_id[0] == 1
@@ -486,9 +484,9 @@ def test_lru_cache_worker_adapter_manager(llama_2_7b_model_extra_embeddings,
assert worker_adapter_manager._adapter_manager.lora_index_to_id[3] == 4
worker_adapter_manager.set_active_adapters([
LoRARequest("6", 6, sql_lora_files),
LoRARequest("7", 7, sql_lora_files),
LoRARequest("8", 8, sql_lora_files)
LoRARequest("6", 6, dummy_lora_files),
LoRARequest("7", 7, dummy_lora_files),
LoRARequest("8", 8, dummy_lora_files)
], mapping)
assert worker_adapter_manager.list_adapters() == {1, 6, 7, 8}
assert worker_adapter_manager._adapter_manager.lora_index_to_id[0] == 1
@@ -499,11 +497,11 @@ def test_lru_cache_worker_adapter_manager(llama_2_7b_model_extra_embeddings,
# Over capacity
with pytest.raises(RuntimeError):
worker_adapter_manager.set_active_adapters([
LoRARequest("10", 10, sql_lora_files),
LoRARequest("11", 11, sql_lora_files),
LoRARequest("12", 12, sql_lora_files),
LoRARequest("13", 13, sql_lora_files),
LoRARequest("14", 14, sql_lora_files)
LoRARequest("10", 10, dummy_lora_files),
LoRARequest("11", 11, dummy_lora_files),
LoRARequest("12", 12, dummy_lora_files),
LoRARequest("13", 13, dummy_lora_files),
LoRARequest("14", 14, dummy_lora_files)
], mapping)
assert worker_adapter_manager.device == device
@@ -512,33 +510,41 @@ def test_lru_cache_worker_adapter_manager(llama_2_7b_model_extra_embeddings,
@pytest.mark.parametrize("device", DEVICES)
def test_worker_adapter_manager(llama_2_7b_model_extra_embeddings,
sql_lora_files, device):
def test_worker_adapter_manager(dist_init, dummy_model_gate_up, device,
tmp_path):
# Should remove every LoRA not specified in the request.
lora_config = LoRAConfig(max_lora_rank=8,
max_cpu_loras=4,
max_loras=4,
lora_dtype=DEFAULT_DTYPE)
worker_adapter_manager = WorkerLoRAManager(
4, 2, llama_2_7b_model_extra_embeddings.unpadded_vocab_size -
4, 2, dummy_model_gate_up.unpadded_vocab_size -
lora_config.lora_extra_vocab_size, lora_config, device,
EMBEDDING_MODULES, EMBEDDING_PADDING_MODULES)
worker_adapter_manager.create_lora_manager(
llama_2_7b_model_extra_embeddings)
worker_adapter_manager.create_lora_manager(dummy_model_gate_up)
dummy_lora_files = f"{tmp_path}/lora_adapter"
os.makedirs(dummy_lora_files, exist_ok=True)
create_peft_lora(
dummy_model_gate_up,
save_dir=dummy_lora_files,
target_modules=["layer1.dense1", "dense2"],
lora_dtype=DEFAULT_DTYPE,
)
mapping = LoRAMapping([], [])
worker_adapter_manager.set_active_adapters([
LoRARequest("1", 1, sql_lora_files),
LoRARequest("2", 2, sql_lora_files)
LoRARequest("1", 1, dummy_lora_files),
LoRARequest("2", 2, dummy_lora_files)
], mapping)
assert worker_adapter_manager.list_adapters() == {1, 2}
assert worker_adapter_manager._adapter_manager.lora_index_to_id[0] == 1
assert worker_adapter_manager._adapter_manager.lora_index_to_id[1] == 2
worker_adapter_manager.set_active_adapters([
LoRARequest("1", 1, sql_lora_files),
LoRARequest("3", 3, sql_lora_files),
LoRARequest("4", 4, sql_lora_files)
LoRARequest("1", 1, dummy_lora_files),
LoRARequest("3", 3, dummy_lora_files),
LoRARequest("4", 4, dummy_lora_files)
], mapping)
assert worker_adapter_manager.list_adapters() == {1, 3, 4}
assert worker_adapter_manager._adapter_manager.lora_index_to_id[0] == 1
@@ -546,9 +552,9 @@ def test_worker_adapter_manager(llama_2_7b_model_extra_embeddings,
assert worker_adapter_manager._adapter_manager.lora_index_to_id[2] == 4
worker_adapter_manager.set_active_adapters([
LoRARequest("1", 1, sql_lora_files),
LoRARequest("2", 2, sql_lora_files),
LoRARequest("5", 5, sql_lora_files)
LoRARequest("1", 1, dummy_lora_files),
LoRARequest("2", 2, dummy_lora_files),
LoRARequest("5", 5, dummy_lora_files)
], mapping)
assert worker_adapter_manager.list_adapters() == {1, 2, 5}
assert worker_adapter_manager._adapter_manager.lora_index_to_id[0] == 1
@@ -556,9 +562,9 @@ def test_worker_adapter_manager(llama_2_7b_model_extra_embeddings,
assert worker_adapter_manager._adapter_manager.lora_index_to_id[2] == 5
worker_adapter_manager.set_active_adapters([
LoRARequest("1", 1, sql_lora_files),
LoRARequest("1", 1, sql_lora_files),
LoRARequest("1", 1, sql_lora_files)
LoRARequest("1", 1, dummy_lora_files),
LoRARequest("1", 1, dummy_lora_files),
LoRARequest("1", 1, dummy_lora_files)
], mapping)
assert worker_adapter_manager.list_adapters() == {1}
assert worker_adapter_manager._adapter_manager.lora_index_to_id[0] == 1
@@ -566,9 +572,9 @@ def test_worker_adapter_manager(llama_2_7b_model_extra_embeddings,
assert worker_adapter_manager._adapter_manager.lora_index_to_id[2] is None
worker_adapter_manager.set_active_adapters([
LoRARequest("6", 6, sql_lora_files),
LoRARequest("7", 7, sql_lora_files),
LoRARequest("8", 8, sql_lora_files)
LoRARequest("6", 6, dummy_lora_files),
LoRARequest("7", 7, dummy_lora_files),
LoRARequest("8", 8, dummy_lora_files)
], mapping)
assert worker_adapter_manager.list_adapters() == {6, 7, 8}
assert worker_adapter_manager._adapter_manager.lora_index_to_id[0] == 8
@@ -578,11 +584,11 @@ def test_worker_adapter_manager(llama_2_7b_model_extra_embeddings,
# Over capacity
with pytest.raises(RuntimeError):
worker_adapter_manager.set_active_adapters([
LoRARequest("10", 10, sql_lora_files),
LoRARequest("11", 11, sql_lora_files),
LoRARequest("12", 12, sql_lora_files),
LoRARequest("13", 13, sql_lora_files),
LoRARequest("14", 14, sql_lora_files)
LoRARequest("10", 10, dummy_lora_files),
LoRARequest("11", 11, dummy_lora_files),
LoRARequest("12", 12, dummy_lora_files),
LoRARequest("13", 13, dummy_lora_files),
LoRARequest("14", 14, dummy_lora_files)
], mapping)
assert worker_adapter_manager.device == device