Convert formatting to use ruff instead of yapf + isort (#26247)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
+65
-48
@@ -6,37 +6,44 @@ from typing import TYPE_CHECKING, Optional, Union
|
||||
|
||||
import huggingface_hub
|
||||
import regex as re
|
||||
from huggingface_hub.utils import (EntryNotFoundError, HfHubHTTPError,
|
||||
HFValidationError, RepositoryNotFoundError)
|
||||
from huggingface_hub.utils import (
|
||||
EntryNotFoundError,
|
||||
HfHubHTTPError,
|
||||
HFValidationError,
|
||||
RepositoryNotFoundError,
|
||||
)
|
||||
from torch import nn
|
||||
from transformers import PretrainedConfig
|
||||
|
||||
from vllm.config.lora import LoRAConfig
|
||||
from vllm.logger import init_logger
|
||||
|
||||
# being imported for _all_lora_classes below
|
||||
# yapf conflicts with isort for this block
|
||||
# yapf: disable
|
||||
from vllm.lora.layers import (BaseLayerWithLoRA, ColumnParallelLinearWithLoRA,
|
||||
ColumnParallelLinearWithShardedLoRA,
|
||||
LogitsProcessorWithLoRA,
|
||||
MergedColumnParallelLinearWithLoRA,
|
||||
MergedColumnParallelLinearWithShardedLoRA,
|
||||
MergedQKVParallelLinearWithLoRA,
|
||||
MergedQKVParallelLinearWithShardedLoRA,
|
||||
QKVParallelLinearWithLoRA,
|
||||
QKVParallelLinearWithShardedLoRA,
|
||||
ReplicatedLinearWithLoRA,
|
||||
RowParallelLinearWithLoRA,
|
||||
RowParallelLinearWithShardedLoRA,
|
||||
VocabParallelEmbeddingWithLoRA)
|
||||
from vllm.lora.layers import (
|
||||
BaseLayerWithLoRA,
|
||||
ColumnParallelLinearWithLoRA,
|
||||
ColumnParallelLinearWithShardedLoRA,
|
||||
LogitsProcessorWithLoRA,
|
||||
MergedColumnParallelLinearWithLoRA,
|
||||
MergedColumnParallelLinearWithShardedLoRA,
|
||||
MergedQKVParallelLinearWithLoRA,
|
||||
MergedQKVParallelLinearWithShardedLoRA,
|
||||
QKVParallelLinearWithLoRA,
|
||||
QKVParallelLinearWithShardedLoRA,
|
||||
ReplicatedLinearWithLoRA,
|
||||
RowParallelLinearWithLoRA,
|
||||
RowParallelLinearWithShardedLoRA,
|
||||
VocabParallelEmbeddingWithLoRA,
|
||||
)
|
||||
from vllm.model_executor.layers.linear import LinearBase
|
||||
|
||||
# yapf: enable
|
||||
|
||||
if TYPE_CHECKING:
|
||||
from vllm.model_executor.layers.logits_processor import LogitsProcessor
|
||||
from vllm.model_executor.layers.vocab_parallel_embedding import (
|
||||
ParallelLMHead)
|
||||
from vllm.model_executor.layers.vocab_parallel_embedding import ParallelLMHead
|
||||
from vllm.model_executor.models.utils import WeightsMapper
|
||||
|
||||
logger = init_logger(__name__)
|
||||
@@ -58,20 +65,23 @@ _all_lora_classes: set[type[BaseLayerWithLoRA]] = {
|
||||
}
|
||||
|
||||
|
||||
def from_layer(layer: nn.Module,
|
||||
max_loras: int,
|
||||
lora_config: LoRAConfig,
|
||||
packed_modules_list: list,
|
||||
model_config: Optional[PretrainedConfig] = None) -> nn.Module:
|
||||
def from_layer(
|
||||
layer: nn.Module,
|
||||
max_loras: int,
|
||||
lora_config: LoRAConfig,
|
||||
packed_modules_list: list,
|
||||
model_config: Optional[PretrainedConfig] = None,
|
||||
) -> nn.Module:
|
||||
for lora_cls in _all_lora_classes:
|
||||
# specifying kwargs so they can be easily accessed in decorator
|
||||
if lora_cls.can_replace_layer(source_layer=layer,
|
||||
lora_config=lora_config,
|
||||
packed_modules_list=packed_modules_list,
|
||||
model_config=model_config):
|
||||
if lora_cls.can_replace_layer(
|
||||
source_layer=layer,
|
||||
lora_config=lora_config,
|
||||
packed_modules_list=packed_modules_list,
|
||||
model_config=model_config,
|
||||
):
|
||||
instance_layer = lora_cls(layer)
|
||||
instance_layer.create_lora_weights(max_loras, lora_config,
|
||||
model_config)
|
||||
instance_layer.create_lora_weights(max_loras, lora_config, model_config)
|
||||
return instance_layer
|
||||
return layer
|
||||
|
||||
@@ -83,15 +93,20 @@ def from_layer_logits_processor(
|
||||
lora_config: LoRAConfig,
|
||||
model_config: Optional[PretrainedConfig] = None,
|
||||
) -> LogitsProcessorWithLoRA:
|
||||
ret = LogitsProcessorWithLoRA(layer, lm_head.embedding_dim,
|
||||
lm_head.weight.dtype, lm_head.weight.device,
|
||||
lm_head.get_sharded_to_full_mapping())
|
||||
ret = LogitsProcessorWithLoRA(
|
||||
layer,
|
||||
lm_head.embedding_dim,
|
||||
lm_head.weight.dtype,
|
||||
lm_head.weight.device,
|
||||
lm_head.get_sharded_to_full_mapping(),
|
||||
)
|
||||
ret.create_lora_weights(max_loras, lora_config, model_config)
|
||||
return ret
|
||||
|
||||
|
||||
def replace_submodule(model: nn.Module, module_name: str,
|
||||
new_module: nn.Module) -> nn.Module:
|
||||
def replace_submodule(
|
||||
model: nn.Module, module_name: str, new_module: nn.Module
|
||||
) -> nn.Module:
|
||||
"""Replace a submodule in a model with a new module."""
|
||||
parent = model.get_submodule(".".join(module_name.split(".")[:-1]))
|
||||
target_name = module_name.split(".")[-1]
|
||||
@@ -100,8 +115,7 @@ def replace_submodule(model: nn.Module, module_name: str,
|
||||
|
||||
|
||||
def parse_fine_tuned_lora_name(
|
||||
name: str,
|
||||
weights_mapper: Optional["WeightsMapper"] = None
|
||||
name: str, weights_mapper: Optional["WeightsMapper"] = None
|
||||
) -> tuple[str, bool, bool]:
|
||||
"""Parse the name of lora weights.
|
||||
|
||||
@@ -134,8 +148,7 @@ def parse_fine_tuned_lora_name(
|
||||
start_index = 2 if name.startswith("base_model.model.") else 0
|
||||
|
||||
parts = name.split(".")
|
||||
if parts[-1] == "weight" and (parts[-2] == "lora_A"
|
||||
or parts[-2] == "lora_B"):
|
||||
if parts[-1] == "weight" and (parts[-2] == "lora_A" or parts[-2] == "lora_B"):
|
||||
new_name = ".".join(parts[start_index:-2])
|
||||
return new_name, parts[-2] == "lora_A", False
|
||||
|
||||
@@ -150,12 +163,13 @@ def parse_fine_tuned_lora_name(
|
||||
raise ValueError(f"{name} is unsupported LoRA weight")
|
||||
|
||||
|
||||
def is_regex_target_modules(load_modules: Union[str, list[str]],
|
||||
expected_lora_modules: list[str]) -> bool:
|
||||
def is_regex_target_modules(
|
||||
load_modules: Union[str, list[str]], expected_lora_modules: list[str]
|
||||
) -> bool:
|
||||
"""
|
||||
PEFT supports passing `target_modules` in the form of regular expressions,
|
||||
such as `model.*(q_proj|k_proj|v_proj)$`. This function is mainly used to
|
||||
determine whether the suffix in the regular expression is present in the
|
||||
PEFT supports passing `target_modules` in the form of regular expressions,
|
||||
such as `model.*(q_proj|k_proj|v_proj)$`. This function is mainly used to
|
||||
determine whether the suffix in the regular expression is present in the
|
||||
`expected_lora_modules`.
|
||||
"""
|
||||
|
||||
@@ -197,7 +211,7 @@ def get_supported_lora_modules(model: nn.Module) -> list[str]:
|
||||
supported_lora_modules.add(name)
|
||||
|
||||
# get all the linear subfixes.
|
||||
if isinstance(module, (LinearBase, )):
|
||||
if isinstance(module, (LinearBase,)):
|
||||
supported_lora_modules.add(name.split(".")[-1])
|
||||
|
||||
return list(supported_lora_modules)
|
||||
@@ -225,7 +239,7 @@ def get_adapter_absolute_path(lora_path: str) -> str:
|
||||
return lora_path
|
||||
|
||||
# If the path starts with ~, expand the user home directory.
|
||||
if lora_path.startswith('~'):
|
||||
if lora_path.startswith("~"):
|
||||
return os.path.expanduser(lora_path)
|
||||
|
||||
# Check if the expanded relative path exists locally.
|
||||
@@ -234,10 +248,13 @@ def get_adapter_absolute_path(lora_path: str) -> str:
|
||||
|
||||
# If the path does not exist locally, assume it's a Hugging Face repo.
|
||||
try:
|
||||
local_snapshot_path = huggingface_hub.snapshot_download(
|
||||
repo_id=lora_path)
|
||||
except (HfHubHTTPError, RepositoryNotFoundError, EntryNotFoundError,
|
||||
HFValidationError):
|
||||
local_snapshot_path = huggingface_hub.snapshot_download(repo_id=lora_path)
|
||||
except (
|
||||
HfHubHTTPError,
|
||||
RepositoryNotFoundError,
|
||||
EntryNotFoundError,
|
||||
HFValidationError,
|
||||
):
|
||||
# Handle errors that may occur during the download
|
||||
# Return original path instead of throwing error here
|
||||
logger.exception("Error downloading the HuggingFace model")
|
||||
|
||||
Reference in New Issue
Block a user