[Misc] Improve LoRA spelling (#13831)
This commit is contained in:
@@ -491,7 +491,7 @@ def test_prefill_schedule_max_lora():
|
||||
lora_path="abc"))
|
||||
scheduler.add_seq_group(seq_group)
|
||||
# Add two more requests to verify lora is prioritized.
|
||||
# 0: Lora, 1: Lora, 2: regular, 3: regular
|
||||
# 0: LoRA, 1: LoRA, 2: regular, 3: regular
|
||||
# In the first iteration, index 0, 2 is scheduled.
|
||||
# If a request is not scheduled because it hits max lora, it is
|
||||
# prioritized. Verify that.
|
||||
|
||||
@@ -26,7 +26,7 @@ def serve_parser():
|
||||
return make_arg_parser(parser)
|
||||
|
||||
|
||||
### Tests for Lora module parsing
|
||||
### Tests for LoRA module parsing
|
||||
def test_valid_key_value_format(serve_parser):
|
||||
# Test old format: name=path
|
||||
args = serve_parser.parse_args([
|
||||
|
||||
@@ -8,8 +8,8 @@ import pytest
|
||||
from vllm.config import ModelConfig
|
||||
from vllm.engine.protocol import EngineClient
|
||||
from vllm.entrypoints.openai.protocol import (ErrorResponse,
|
||||
LoadLoraAdapterRequest,
|
||||
UnloadLoraAdapterRequest)
|
||||
LoadLoRAAdapterRequest,
|
||||
UnloadLoRAAdapterRequest)
|
||||
from vllm.entrypoints.openai.serving_models import (BaseModelPath,
|
||||
OpenAIServingModels)
|
||||
from vllm.lora.request import LoRARequest
|
||||
@@ -51,7 +51,7 @@ async def test_serving_model_name():
|
||||
@pytest.mark.asyncio
|
||||
async def test_load_lora_adapter_success():
|
||||
serving_models = await _async_serving_models_init()
|
||||
request = LoadLoraAdapterRequest(lora_name="adapter",
|
||||
request = LoadLoRAAdapterRequest(lora_name="adapter",
|
||||
lora_path="/path/to/adapter2")
|
||||
response = await serving_models.load_lora_adapter(request)
|
||||
assert response == LORA_LOADING_SUCCESS_MESSAGE.format(lora_name='adapter')
|
||||
@@ -62,7 +62,7 @@ async def test_load_lora_adapter_success():
|
||||
@pytest.mark.asyncio
|
||||
async def test_load_lora_adapter_missing_fields():
|
||||
serving_models = await _async_serving_models_init()
|
||||
request = LoadLoraAdapterRequest(lora_name="", lora_path="")
|
||||
request = LoadLoRAAdapterRequest(lora_name="", lora_path="")
|
||||
response = await serving_models.load_lora_adapter(request)
|
||||
assert isinstance(response, ErrorResponse)
|
||||
assert response.type == "InvalidUserInput"
|
||||
@@ -72,14 +72,14 @@ async def test_load_lora_adapter_missing_fields():
|
||||
@pytest.mark.asyncio
|
||||
async def test_load_lora_adapter_duplicate():
|
||||
serving_models = await _async_serving_models_init()
|
||||
request = LoadLoraAdapterRequest(lora_name="adapter1",
|
||||
request = LoadLoRAAdapterRequest(lora_name="adapter1",
|
||||
lora_path="/path/to/adapter1")
|
||||
response = await serving_models.load_lora_adapter(request)
|
||||
assert response == LORA_LOADING_SUCCESS_MESSAGE.format(
|
||||
lora_name='adapter1')
|
||||
assert len(serving_models.lora_requests) == 1
|
||||
|
||||
request = LoadLoraAdapterRequest(lora_name="adapter1",
|
||||
request = LoadLoRAAdapterRequest(lora_name="adapter1",
|
||||
lora_path="/path/to/adapter1")
|
||||
response = await serving_models.load_lora_adapter(request)
|
||||
assert isinstance(response, ErrorResponse)
|
||||
@@ -91,12 +91,12 @@ async def test_load_lora_adapter_duplicate():
|
||||
@pytest.mark.asyncio
|
||||
async def test_unload_lora_adapter_success():
|
||||
serving_models = await _async_serving_models_init()
|
||||
request = LoadLoraAdapterRequest(lora_name="adapter1",
|
||||
request = LoadLoRAAdapterRequest(lora_name="adapter1",
|
||||
lora_path="/path/to/adapter1")
|
||||
response = await serving_models.load_lora_adapter(request)
|
||||
assert len(serving_models.lora_requests) == 1
|
||||
|
||||
request = UnloadLoraAdapterRequest(lora_name="adapter1")
|
||||
request = UnloadLoRAAdapterRequest(lora_name="adapter1")
|
||||
response = await serving_models.unload_lora_adapter(request)
|
||||
assert response == LORA_UNLOADING_SUCCESS_MESSAGE.format(
|
||||
lora_name='adapter1')
|
||||
@@ -106,7 +106,7 @@ async def test_unload_lora_adapter_success():
|
||||
@pytest.mark.asyncio
|
||||
async def test_unload_lora_adapter_missing_fields():
|
||||
serving_models = await _async_serving_models_init()
|
||||
request = UnloadLoraAdapterRequest(lora_name="", lora_int_id=None)
|
||||
request = UnloadLoRAAdapterRequest(lora_name="", lora_int_id=None)
|
||||
response = await serving_models.unload_lora_adapter(request)
|
||||
assert isinstance(response, ErrorResponse)
|
||||
assert response.type == "InvalidUserInput"
|
||||
@@ -116,7 +116,7 @@ async def test_unload_lora_adapter_missing_fields():
|
||||
@pytest.mark.asyncio
|
||||
async def test_unload_lora_adapter_not_found():
|
||||
serving_models = await _async_serving_models_init()
|
||||
request = UnloadLoraAdapterRequest(lora_name="nonexistent_adapter")
|
||||
request = UnloadLoRAAdapterRequest(lora_name="nonexistent_adapter")
|
||||
response = await serving_models.unload_lora_adapter(request)
|
||||
assert isinstance(response, ErrorResponse)
|
||||
assert response.type == "NotFoundError"
|
||||
|
||||
@@ -14,16 +14,16 @@ from vllm.config import LoRAConfig
|
||||
from vllm.lora.fully_sharded_layers import (
|
||||
ColumnParallelLinearWithShardedLoRA,
|
||||
MergedColumnParallelLinearWithShardedLoRA,
|
||||
MergedQKVParallelLinearWithShardedLora, QKVParallelLinearWithShardedLora,
|
||||
MergedQKVParallelLinearWithShardedLoRA, QKVParallelLinearWithShardedLoRA,
|
||||
RowParallelLinearWithShardedLoRA)
|
||||
# yapf conflicts with isort for this block
|
||||
# yapf: disable
|
||||
from vllm.lora.layers import (BaseLayerWithLoRA, ColumnParallelLinearWithLoRA,
|
||||
LinearScalingRotaryEmbeddingWithLora,
|
||||
LinearScalingRotaryEmbeddingWithLoRA,
|
||||
LogitsProcessorWithLoRA, LoRAMapping,
|
||||
MergedColumnParallelLinearWithLoRA,
|
||||
MergedQKVParallelLinearWithLora,
|
||||
QKVParallelLinearWithLora,
|
||||
MergedQKVParallelLinearWithLoRA,
|
||||
QKVParallelLinearWithLoRA,
|
||||
ReplicatedLinearWithLoRA,
|
||||
RowParallelLinearWithLoRA,
|
||||
VocabParallelEmbeddingWithLoRA)
|
||||
@@ -866,9 +866,9 @@ def test_column_parallel_packed(dist_init, num_loras, repeats, fully_shard,
|
||||
bias=False,
|
||||
params_dtype=torch.float16)
|
||||
linear.weight.data = torch.rand_like(linear.weight.data)
|
||||
lora_linear = (MergedQKVParallelLinearWithLora(linear)
|
||||
lora_linear = (MergedQKVParallelLinearWithLoRA(linear)
|
||||
if not fully_shard else
|
||||
MergedQKVParallelLinearWithShardedLora(linear))
|
||||
MergedQKVParallelLinearWithShardedLoRA(linear))
|
||||
else:
|
||||
linear = QKVParallelLinear(4096,
|
||||
64,
|
||||
@@ -876,9 +876,9 @@ def test_column_parallel_packed(dist_init, num_loras, repeats, fully_shard,
|
||||
bias=False,
|
||||
params_dtype=torch.float16)
|
||||
linear.weight.data = torch.rand_like(linear.weight.data)
|
||||
lora_linear = QKVParallelLinearWithLora(
|
||||
lora_linear = QKVParallelLinearWithLoRA(
|
||||
linear
|
||||
) if not fully_shard else QKVParallelLinearWithShardedLora(linear)
|
||||
) if not fully_shard else QKVParallelLinearWithShardedLoRA(linear)
|
||||
|
||||
@dataclass
|
||||
class FakeConfig:
|
||||
@@ -1024,7 +1024,7 @@ def test_rotary_embedding_long_context(dist_init, num_loras, device,
|
||||
base,
|
||||
is_neox_style,
|
||||
)
|
||||
lora_rope = LinearScalingRotaryEmbeddingWithLora(rope)
|
||||
lora_rope = LinearScalingRotaryEmbeddingWithLoRA(rope)
|
||||
lora_rope.set_mapping(punica_wrapper)
|
||||
lora_rope.create_lora_weights(max_loras, lora_config)
|
||||
linear_rope = get_rope(head_size, rotary_dim, max_position, base,
|
||||
|
||||
@@ -8,7 +8,7 @@ import pytest
|
||||
|
||||
import vllm
|
||||
from vllm import SamplingParams
|
||||
from vllm.lora.layers import LinearScalingRotaryEmbeddingWithLora
|
||||
from vllm.lora.layers import LinearScalingRotaryEmbeddingWithLoRA
|
||||
from vllm.lora.request import LoRARequest
|
||||
from vllm.model_executor.layers.rotary_embedding import (
|
||||
LinearScalingRotaryEmbedding)
|
||||
@@ -151,7 +151,7 @@ def test_rotary_emb_replaced(dist_init):
|
||||
if "rotary_emb" in module_name:
|
||||
if "base_layer" not in module_name:
|
||||
rotary_emb_count += 1
|
||||
assert isinstance(module, LinearScalingRotaryEmbeddingWithLora)
|
||||
assert isinstance(module, LinearScalingRotaryEmbeddingWithLoRA)
|
||||
else:
|
||||
assert isinstance(module, LinearScalingRotaryEmbedding)
|
||||
# Llama 2 has 32 layers.
|
||||
|
||||
Reference in New Issue
Block a user