Migrate linter from pylint to ruff (#1665)
This commit is contained in:
@@ -350,7 +350,7 @@ class Scheduler:
|
||||
elif preemption_mode == PreemptionMode.SWAP:
|
||||
self._preempt_by_swap(seq_group, blocks_to_swap_out)
|
||||
else:
|
||||
assert False, "Invalid preemption mode."
|
||||
raise AssertionError("Invalid preemption mode.")
|
||||
|
||||
def _preempt_by_recompute(
|
||||
self,
|
||||
|
||||
@@ -125,7 +125,7 @@ class LLMEngine:
|
||||
def _init_workers(self, distributed_init_method: str):
|
||||
# Lazy import the Worker to avoid importing torch.cuda/xformers
|
||||
# before CUDA_VISIBLE_DEVICES is set in the Worker
|
||||
from vllm.worker.worker import Worker # pylint: disable=import-outside-toplevel
|
||||
from vllm.worker.worker import Worker
|
||||
|
||||
assert self.parallel_config.world_size == 1, (
|
||||
"Ray is required if parallel_config.world_size > 1.")
|
||||
@@ -148,7 +148,7 @@ class LLMEngine:
|
||||
**ray_remote_kwargs):
|
||||
# Lazy import the Worker to avoid importing torch.cuda/xformers
|
||||
# before CUDA_VISIBLE_DEVICES is set in the Worker
|
||||
from vllm.worker.worker import Worker # pylint: disable=import-outside-toplevel
|
||||
from vllm.worker.worker import Worker
|
||||
|
||||
self.workers: List[Worker] = []
|
||||
for bundle in placement_group.bundle_specs:
|
||||
|
||||
@@ -16,7 +16,6 @@ try:
|
||||
|
||||
def __init__(self, init_cached_hf_modules=False) -> None:
|
||||
if init_cached_hf_modules:
|
||||
# pylint: disable=import-outside-toplevel
|
||||
from transformers.dynamic_module_utils import init_hf_modules
|
||||
init_hf_modules()
|
||||
self.worker = None
|
||||
@@ -37,7 +36,7 @@ except ImportError as e:
|
||||
"`pip install ray pandas pyarrow`.")
|
||||
ray = None
|
||||
TorchDistributedWorker = None
|
||||
RayWorker = None # pylint: disable=invalid-name
|
||||
RayWorker = None
|
||||
|
||||
if TYPE_CHECKING:
|
||||
from ray.util.placement_group import PlacementGroup
|
||||
|
||||
+8
-12
@@ -134,25 +134,21 @@ class LLM:
|
||||
if isinstance(prompts, str):
|
||||
# Convert a single prompt to a list.
|
||||
prompts = [prompts]
|
||||
if prompts is not None and prompt_token_ids is not None:
|
||||
if len(prompts) != len(prompt_token_ids):
|
||||
raise ValueError("The lengths of prompts and prompt_token_ids "
|
||||
"must be the same.")
|
||||
if (prompts is not None and prompt_token_ids is not None
|
||||
and len(prompts) != len(prompt_token_ids)):
|
||||
raise ValueError("The lengths of prompts and prompt_token_ids "
|
||||
"must be the same.")
|
||||
if sampling_params is None:
|
||||
# Use default sampling params.
|
||||
sampling_params = SamplingParams()
|
||||
|
||||
# Add requests to the engine.
|
||||
if prompts is not None:
|
||||
num_requests = len(prompts)
|
||||
else:
|
||||
num_requests = len(prompt_token_ids)
|
||||
num_requests = len(prompts) if prompts is not None else len(
|
||||
prompt_token_ids)
|
||||
for i in range(num_requests):
|
||||
prompt = prompts[i] if prompts is not None else None
|
||||
if prompt_token_ids is None:
|
||||
token_ids = None
|
||||
else:
|
||||
token_ids = prompt_token_ids[i]
|
||||
token_ids = None if prompt_token_ids is None else prompt_token_ids[
|
||||
i]
|
||||
self._add_request(prompt, sampling_params, token_ids)
|
||||
return self._run_engine(use_tqdm)
|
||||
|
||||
|
||||
@@ -55,7 +55,7 @@ def create_error_response(status_code: HTTPStatus,
|
||||
|
||||
|
||||
@app.exception_handler(RequestValidationError)
|
||||
async def validation_exception_handler(request, exc): # pylint: disable=unused-argument
|
||||
async def validation_exception_handler(_, exc):
|
||||
return create_error_response(HTTPStatus.BAD_REQUEST, str(exc))
|
||||
|
||||
|
||||
@@ -124,10 +124,8 @@ async def check_length(
|
||||
assert (not (prompt is None and prompt_ids is None)
|
||||
and not (prompt is not None and prompt_ids is not None)
|
||||
), "Either prompt or prompt_ids should be provided."
|
||||
if prompt_ids is not None:
|
||||
input_ids = prompt_ids
|
||||
else:
|
||||
input_ids = tokenizer(prompt).input_ids
|
||||
input_ids = prompt_ids if prompt_ids is not None else tokenizer(
|
||||
prompt).input_ids
|
||||
token_num = len(input_ids)
|
||||
|
||||
if request.max_tokens is None:
|
||||
|
||||
@@ -84,15 +84,14 @@ def get_act_fn(
|
||||
f"Activation function {act_fn_name!r} is not supported.")
|
||||
|
||||
act_fn = _ACTIVATION_REGISTRY[act_fn_name]
|
||||
if quant_config is not None:
|
||||
if act_fn_name in quant_config.get_scaled_act_names():
|
||||
if intermediate_size is None:
|
||||
raise ValueError(
|
||||
"intermediate_size must be specified for scaled "
|
||||
"activation functions.")
|
||||
return ScaledActivation(
|
||||
act_fn,
|
||||
intermediate_size,
|
||||
params_dtype=torch.get_default_dtype(),
|
||||
)
|
||||
if quant_config is not None and act_fn_name in quant_config.get_scaled_act_names(
|
||||
):
|
||||
if intermediate_size is None:
|
||||
raise ValueError("intermediate_size must be specified for scaled "
|
||||
"activation functions.")
|
||||
return ScaledActivation(
|
||||
act_fn,
|
||||
intermediate_size,
|
||||
params_dtype=torch.get_default_dtype(),
|
||||
)
|
||||
return act_fn
|
||||
|
||||
@@ -18,7 +18,6 @@ _PARTITION_SIZE = 512
|
||||
|
||||
|
||||
class PagedAttention(nn.Module):
|
||||
# pylint: disable=line-too-long
|
||||
"""GPT-style multi-head PagedAttention.
|
||||
|
||||
This class takes query, key, and value tensors as input. The input tensors
|
||||
|
||||
@@ -50,7 +50,7 @@ class AWQConfig(QuantizationConfig):
|
||||
def get_config_filenames() -> List[str]:
|
||||
return [
|
||||
"quant_config.json", # E.g., casperhansen/vicuna-7b-v1.5-awq
|
||||
"quantize_config.json", # E.g., abhinavkulkarni/mosaicml-mpt-7b-instruct-w4-g128-awq # pylint: disable=line-too-long
|
||||
"quantize_config.json", # E.g., abhinavkulkarni/mosaicml-mpt-7b-instruct-w4-g128-awq
|
||||
]
|
||||
|
||||
@classmethod
|
||||
|
||||
@@ -7,7 +7,7 @@ import torch.nn as nn
|
||||
from transformers import PretrainedConfig
|
||||
|
||||
from vllm.config import ModelConfig
|
||||
from vllm.model_executor.models import * # pylint: disable=wildcard-import
|
||||
from vllm.model_executor.models import *
|
||||
from vllm.model_executor.weight_utils import (get_quant_config,
|
||||
initialize_dummy_weights)
|
||||
|
||||
|
||||
@@ -261,10 +261,7 @@ class AquilaModel(nn.Module):
|
||||
) -> torch.Tensor:
|
||||
hidden_states = self.embed_tokens(input_ids)
|
||||
for i in range(len(self.layers)):
|
||||
if cache_events is None:
|
||||
cache_event = None
|
||||
else:
|
||||
cache_event = cache_events[i]
|
||||
cache_event = None if cache_events is None else cache_events[i]
|
||||
layer = self.layers[i]
|
||||
hidden_states = layer(
|
||||
positions,
|
||||
|
||||
@@ -281,10 +281,7 @@ class BaiChuanModel(nn.Module):
|
||||
hidden_states = self.embed_tokens(input_ids)
|
||||
residual = None
|
||||
for i in range(len(self.layers)):
|
||||
if cache_events is None:
|
||||
cache_event = None
|
||||
else:
|
||||
cache_event = cache_events[i]
|
||||
cache_event = None if cache_events is None else cache_events[i]
|
||||
layer = self.layers[i]
|
||||
hidden_states, residual = layer(
|
||||
positions,
|
||||
|
||||
@@ -256,10 +256,7 @@ class BloomModel(nn.Module):
|
||||
hidden_states = self.word_embeddings(input_ids)
|
||||
hidden_states = self.word_embeddings_layernorm(hidden_states)
|
||||
for i in range(len(self.h)):
|
||||
if cache_events is None:
|
||||
cache_event = None
|
||||
else:
|
||||
cache_event = cache_events[i]
|
||||
cache_event = None if cache_events is None else cache_events[i]
|
||||
layer = self.h[i]
|
||||
hidden_states = layer(
|
||||
position_ids,
|
||||
|
||||
@@ -269,10 +269,7 @@ class GLMTransformer(nn.Module):
|
||||
cache_events: Optional[List[torch.cuda.Event]],
|
||||
) -> torch.Tensor:
|
||||
for i in range(self.num_layers):
|
||||
if cache_events is None:
|
||||
cache_event = None
|
||||
else:
|
||||
cache_event = cache_events[i]
|
||||
cache_event = None if cache_events is None else cache_events[i]
|
||||
layer = self.layers[i]
|
||||
hidden_states = layer(
|
||||
hidden_states=hidden_states,
|
||||
|
||||
@@ -353,10 +353,7 @@ class FalconModel(nn.Module):
|
||||
) -> torch.Tensor:
|
||||
hidden_states = self.word_embeddings(input_ids)
|
||||
for i in range(len(self.h)):
|
||||
if cache_events is None:
|
||||
cache_event = None
|
||||
else:
|
||||
cache_event = cache_events[i]
|
||||
cache_event = None if cache_events is None else cache_events[i]
|
||||
layer = self.h[i]
|
||||
hidden_states = layer(
|
||||
positions,
|
||||
|
||||
@@ -206,10 +206,7 @@ class GPT2Model(nn.Module):
|
||||
hidden_states = inputs_embeds + position_embeds
|
||||
|
||||
for i in range(len(self.h)):
|
||||
if cache_events is None:
|
||||
cache_event = None
|
||||
else:
|
||||
cache_event = cache_events[i]
|
||||
cache_event = None if cache_events is None else cache_events[i]
|
||||
layer = self.h[i]
|
||||
hidden_states = layer(hidden_states, kv_caches[i], input_metadata,
|
||||
cache_event)
|
||||
|
||||
@@ -225,10 +225,7 @@ class GPTBigCodeModel(nn.Module):
|
||||
hidden_states = inputs_embeds + position_embeds
|
||||
|
||||
for i in range(len(self.h)):
|
||||
if cache_events is None:
|
||||
cache_event = None
|
||||
else:
|
||||
cache_event = cache_events[i]
|
||||
cache_event = None if cache_events is None else cache_events[i]
|
||||
layer = self.h[i]
|
||||
hidden_states = layer(hidden_states, kv_caches[i], input_metadata,
|
||||
cache_event)
|
||||
|
||||
@@ -147,10 +147,7 @@ class GPTJBlock(nn.Module):
|
||||
linear_method: Optional[LinearMethodBase] = None,
|
||||
):
|
||||
super().__init__()
|
||||
if config.n_inner is None:
|
||||
inner_dim = 4 * config.n_embd
|
||||
else:
|
||||
inner_dim = config.n_inner
|
||||
inner_dim = 4 * config.n_embd if config.n_inner is None else config.n_inner
|
||||
self.ln_1 = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
|
||||
self.attn = GPTJAttention(config, linear_method)
|
||||
self.mlp = GPTJMLP(inner_dim, config, linear_method)
|
||||
@@ -205,10 +202,7 @@ class GPTJModel(nn.Module):
|
||||
) -> torch.Tensor:
|
||||
hidden_states = self.wte(input_ids)
|
||||
for i in range(len(self.h)):
|
||||
if cache_events is None:
|
||||
cache_event = None
|
||||
else:
|
||||
cache_event = cache_events[i]
|
||||
cache_event = None if cache_events is None else cache_events[i]
|
||||
layer = self.h[i]
|
||||
hidden_states = layer(
|
||||
position_ids,
|
||||
|
||||
@@ -216,10 +216,7 @@ class GPTNeoXModel(nn.Module):
|
||||
) -> torch.Tensor:
|
||||
hidden_states = self.embed_in(input_ids)
|
||||
for i in range(len(self.layers)):
|
||||
if cache_events is None:
|
||||
cache_event = None
|
||||
else:
|
||||
cache_event = cache_events[i]
|
||||
cache_event = None if cache_events is None else cache_events[i]
|
||||
layer = self.layers[i]
|
||||
hidden_states = layer(
|
||||
position_ids,
|
||||
|
||||
@@ -213,10 +213,7 @@ class InternLMModel(nn.Module):
|
||||
hidden_states = self.embed_tokens(input_ids)
|
||||
residual = None
|
||||
for i in range(len(self.layers)):
|
||||
if cache_events is None:
|
||||
cache_event = None
|
||||
else:
|
||||
cache_event = cache_events[i]
|
||||
cache_event = None if cache_events is None else cache_events[i]
|
||||
layer = self.layers[i]
|
||||
hidden_states, residual = layer(
|
||||
positions,
|
||||
|
||||
@@ -253,10 +253,7 @@ class LlamaModel(nn.Module):
|
||||
hidden_states = self.embed_tokens(input_ids)
|
||||
residual = None
|
||||
for i in range(len(self.layers)):
|
||||
if cache_events is None:
|
||||
cache_event = None
|
||||
else:
|
||||
cache_event = cache_events[i]
|
||||
cache_event = None if cache_events is None else cache_events[i]
|
||||
layer = self.layers[i]
|
||||
hidden_states, residual = layer(
|
||||
positions,
|
||||
|
||||
@@ -248,10 +248,7 @@ class MistralModel(nn.Module):
|
||||
hidden_states = self.embed_tokens(input_ids)
|
||||
residual = None
|
||||
for i in range(len(self.layers)):
|
||||
if cache_events is None:
|
||||
cache_event = None
|
||||
else:
|
||||
cache_event = cache_events[i]
|
||||
cache_event = None if cache_events is None else cache_events[i]
|
||||
layer = self.layers[i]
|
||||
hidden_states, residual = layer(
|
||||
positions,
|
||||
|
||||
@@ -203,10 +203,10 @@ class MPTModel(nn.Module):
|
||||
self.norm_f = nn.LayerNorm(config.d_model)
|
||||
if config.no_bias:
|
||||
for module in self.modules():
|
||||
if hasattr(module, "bias"):
|
||||
if isinstance(module.bias, nn.Parameter):
|
||||
# Remove the bias term in Linear and LayerNorm.
|
||||
module.register_parameter("bias", None)
|
||||
if hasattr(module, "bias") and isinstance(
|
||||
module.bias, nn.Parameter):
|
||||
# Remove the bias term in Linear and LayerNorm.
|
||||
module.register_parameter("bias", None)
|
||||
|
||||
def forward(
|
||||
self,
|
||||
@@ -218,10 +218,7 @@ class MPTModel(nn.Module):
|
||||
) -> torch.Tensor:
|
||||
hidden_states = self.wte(input_ids)
|
||||
for i in range(len(self.blocks)):
|
||||
if cache_events is None:
|
||||
cache_event = None
|
||||
else:
|
||||
cache_event = cache_events[i]
|
||||
cache_event = None if cache_events is None else cache_events[i]
|
||||
block = self.blocks[i]
|
||||
hidden_states = block(
|
||||
position_ids,
|
||||
|
||||
@@ -257,10 +257,7 @@ class OPTDecoder(nn.Module):
|
||||
hidden_states = inputs_embeds + pos_embeds
|
||||
|
||||
for i in range(len(self.layers)):
|
||||
if cache_events is None:
|
||||
cache_event = None
|
||||
else:
|
||||
cache_event = cache_events[i]
|
||||
cache_event = None if cache_events is None else cache_events[i]
|
||||
layer = self.layers[i]
|
||||
hidden_states = layer(hidden_states, kv_caches[i], input_metadata,
|
||||
cache_event)
|
||||
|
||||
@@ -258,10 +258,7 @@ class PhiModel(nn.Module):
|
||||
) -> SamplerOutput:
|
||||
hidden_states = self.embd(input_ids)
|
||||
for i in range(self.config.num_hidden_layers):
|
||||
if cache_events is None:
|
||||
cache_event = None
|
||||
else:
|
||||
cache_event = cache_events[i]
|
||||
cache_event = None if cache_events is None else cache_events[i]
|
||||
layer = self.h[i]
|
||||
hidden_states = layer(
|
||||
positions,
|
||||
|
||||
@@ -213,10 +213,7 @@ class QWenModel(nn.Module):
|
||||
hidden_states = self.wte(input_ids)
|
||||
residual = None
|
||||
for i in range(len(self.h)):
|
||||
if cache_events is None:
|
||||
cache_event = None
|
||||
else:
|
||||
cache_event = cache_events[i]
|
||||
cache_event = None if cache_events is None else cache_events[i]
|
||||
layer = self.h[i]
|
||||
hidden_states, residual = layer(
|
||||
positions,
|
||||
|
||||
@@ -249,10 +249,7 @@ class YiModel(nn.Module):
|
||||
hidden_states = self.embed_tokens(input_ids)
|
||||
residual = None
|
||||
for i in range(len(self.layers)):
|
||||
if cache_events is None:
|
||||
cache_event = None
|
||||
else:
|
||||
cache_event = cache_events[i]
|
||||
cache_event = None if cache_events is None else cache_events[i]
|
||||
layer = self.layers[i]
|
||||
hidden_states, residual = layer(
|
||||
positions,
|
||||
|
||||
@@ -131,11 +131,9 @@ def prepare_hf_model_weights(
|
||||
) -> Tuple[str, List[str], bool]:
|
||||
# Download model weights from huggingface.
|
||||
is_local = os.path.isdir(model_name_or_path)
|
||||
if use_safetensors:
|
||||
allow_patterns = ["*.safetensors"]
|
||||
else:
|
||||
# Some quantized models use .pt files for storing the weights.
|
||||
allow_patterns = ["*.bin", "*.pt"]
|
||||
# Some quantized models use .pt files for storing the weights.
|
||||
allow_patterns = ["*.safetensors"
|
||||
] if use_safetensors else ["*.bin", "*.pt"]
|
||||
if not is_local:
|
||||
# Use file lock to prevent multiple processes from
|
||||
# downloading the same model weights at the same time.
|
||||
@@ -242,7 +240,7 @@ def hf_model_weights_iterator(
|
||||
elif use_safetensors:
|
||||
for st_file in hf_weights_files:
|
||||
with safe_open(st_file, framework="pt") as f:
|
||||
for name in f.keys():
|
||||
for name in f:
|
||||
param = f.get_tensor(name)
|
||||
yield name, param
|
||||
else:
|
||||
|
||||
@@ -2,7 +2,7 @@ from typing import Optional
|
||||
|
||||
from transformers import AutoConfig, PretrainedConfig
|
||||
|
||||
from vllm.transformers_utils.configs import * # pylint: disable=wildcard-import
|
||||
from vllm.transformers_utils.configs import *
|
||||
|
||||
_CONFIG_REGISTRY = {
|
||||
"aquila": AquilaConfig,
|
||||
|
||||
@@ -62,7 +62,6 @@ class MPTConfig(PretrainedConfig):
|
||||
fc_type: str = 'torch',
|
||||
verbose: Optional[int] = None,
|
||||
**kwargs: Any):
|
||||
# pylint: disable=line-too-long
|
||||
"""The MPT configuration class.
|
||||
Args:
|
||||
d_model (int): The size of the embedding dimension of the model.
|
||||
@@ -139,10 +138,10 @@ class MPTConfig(PretrainedConfig):
|
||||
self.init_config = init_config
|
||||
self.fc_type = fc_type
|
||||
if verbose is not None:
|
||||
warnings.warn(
|
||||
DeprecationWarning(
|
||||
'verbose argument for MPTConfig is now ignored and will be removed. Use python_log_level instead.'
|
||||
))
|
||||
warnings.warn(DeprecationWarning(
|
||||
'verbose argument for MPTConfig is now ignored and will be removed. Use python_log_level instead.'
|
||||
),
|
||||
stacklevel=2)
|
||||
if 'name' in kwargs:
|
||||
del kwargs['name']
|
||||
if 'loss_fn' in kwargs:
|
||||
@@ -150,8 +149,8 @@ class MPTConfig(PretrainedConfig):
|
||||
if self.attn_config.get('alibi', False):
|
||||
self.learned_pos_emb = False
|
||||
warnings.warn(
|
||||
f'alibi is turned on, setting `learned_pos_emb` to {self.learned_pos_emb}`'
|
||||
)
|
||||
f'alibi is turned on, setting `learned_pos_emb` to {self.learned_pos_emb}`',
|
||||
stacklevel=2)
|
||||
super().__init__(**kwargs)
|
||||
self._validate_config()
|
||||
|
||||
@@ -211,7 +210,8 @@ class MPTConfig(PretrainedConfig):
|
||||
)
|
||||
if not self.learned_pos_emb and (not self.attn_config['alibi']):
|
||||
warnings.warn(
|
||||
'Positional information not being provided to the model.')
|
||||
'Positional information not being provided to the model.',
|
||||
stacklevel=2)
|
||||
if self.fc_type == 'te' or self.ffn_config['ffn_type'] == 'te_ln_mlp':
|
||||
try:
|
||||
# pylint: disable=import-outside-toplevel
|
||||
|
||||
+1
-1
@@ -30,7 +30,7 @@ class Counter:
|
||||
def get_max_shared_memory_bytes(gpu: int = 0) -> int:
|
||||
"""Returns the maximum shared memory per thread block in bytes."""
|
||||
# https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html
|
||||
cudaDevAttrMaxSharedMemoryPerBlockOptin = 97 # pylint: disable=invalid-name
|
||||
cudaDevAttrMaxSharedMemoryPerBlockOptin = 97
|
||||
max_shared_mem = cuda_utils.get_device_attribute(
|
||||
cudaDevAttrMaxSharedMemoryPerBlockOptin, gpu)
|
||||
return int(max_shared_mem)
|
||||
|
||||
@@ -350,10 +350,7 @@ class Worker:
|
||||
self.cache_engine.copy(blocks_to_copy)
|
||||
issued_cache_op = True
|
||||
|
||||
if issued_cache_op:
|
||||
cache_events = self.cache_events
|
||||
else:
|
||||
cache_events = None
|
||||
cache_events = self.cache_events if issued_cache_op else None
|
||||
|
||||
# If there is no input, we don't need to execute the model.
|
||||
if not seq_group_metadata_list:
|
||||
|
||||
Reference in New Issue
Block a user