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

Convert formatting to use ruff instead of yapf + isort (#26247)

Browse Source 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2025-10-05 15:06:22 +01:00
committed by GitHub
parent 17edd8a807
commit d6953beb91
1508 changed files with 115244 additions and 94146 deletions
Download Patch File Download Diff File Expand all files Collapse all files

156
vllm/model_executor/models/gpt2.py
View File

@@ -19,6 +19,7 @@
# See the License for the specific language governing permissions and
# limitations under the License.
"""Inference-only GPT-2 model compatible with HuggingFace weights."""
from collections.abc import Iterable
from itertools import islice
from typing import Optional, Union
@@ -31,27 +32,36 @@ from vllm.attention import Attention
from vllm.compilation.decorators import support_torch_compile
from vllm.config import CacheConfig, VllmConfig
from vllm.distributed.parallel_state import (
get_pp_group, get_tensor_model_parallel_world_size)
get_pp_group,
get_tensor_model_parallel_world_size,
)
from vllm.model_executor.layers.activation import get_act_fn
from vllm.model_executor.layers.linear import (ColumnParallelLinear,
QKVParallelLinear,
RowParallelLinear)
from vllm.model_executor.layers.linear import (
ColumnParallelLinear,
QKVParallelLinear,
RowParallelLinear,
)
from vllm.model_executor.layers.logits_processor import LogitsProcessor
from vllm.model_executor.layers.quantization import QuantizationConfig
from vllm.model_executor.layers.vocab_parallel_embedding import (
ParallelLMHead, VocabParallelEmbedding)
ParallelLMHead,
VocabParallelEmbedding,
)
from vllm.model_executor.model_loader.weight_utils import default_weight_loader
from vllm.sequence import IntermediateTensors
from ..layers.pooler import DispatchPooler, Pooler
from .interfaces import SupportsPP
from .utils import (AutoWeightsLoader, is_pp_missing_parameter,
make_empty_intermediate_tensors_factory, make_layers,
maybe_prefix)
from .utils import (
AutoWeightsLoader,
is_pp_missing_parameter,
make_empty_intermediate_tensors_factory,
make_layers,
maybe_prefix,
)
class GPT2Attention(nn.Module):
def __init__(
self,
config: GPT2Config,
@@ -62,8 +72,7 @@ class GPT2Attention(nn.Module):
super().__init__()
self.hidden_size = config.hidden_size
total_num_heads = config.num_attention_heads
tensor_model_parallel_world_size = (
get_tensor_model_parallel_world_size())
tensor_model_parallel_world_size = get_tensor_model_parallel_world_size()
assert total_num_heads % tensor_model_parallel_world_size == 0
self.num_heads = total_num_heads // tensor_model_parallel_world_size
self.head_dim = self.hidden_size // total_num_heads
@@ -84,12 +93,14 @@ class GPT2Attention(nn.Module):
quant_config=quant_config,
prefix=f"{prefix}.c_proj",
)
self.attn = Attention(self.num_heads,
self.head_dim,
scale=self.scale,
cache_config=cache_config,
quant_config=quant_config,
prefix=f"{prefix}.attn")
self.attn = Attention(
self.num_heads,
self.head_dim,
scale=self.scale,
cache_config=cache_config,
quant_config=quant_config,
prefix=f"{prefix}.attn",
)
def forward(
self,
@@ -103,7 +114,6 @@ class GPT2Attention(nn.Module):
class GPT2MLP(nn.Module):
def __init__(
self,
intermediate_size: int,
@@ -137,7 +147,6 @@ class GPT2MLP(nn.Module):
class GPT2Block(nn.Module):
def __init__(
self,
config: GPT2Config,
@@ -147,19 +156,14 @@ class GPT2Block(nn.Module):
):
super().__init__()
hidden_size = config.hidden_size
inner_dim = (config.n_inner if config.n_inner is not None else 4 *
hidden_size)
inner_dim = config.n_inner if config.n_inner is not None else 4 * hidden_size
self.ln_1 = nn.LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
self.attn = GPT2Attention(config,
cache_config,
quant_config,
prefix=f"{prefix}.attn")
self.attn = GPT2Attention(
config, cache_config, quant_config, prefix=f"{prefix}.attn"
)
self.ln_2 = nn.LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
self.mlp = GPT2MLP(inner_dim,
config,
quant_config,
prefix=f"{prefix}.mlp")
self.mlp = GPT2MLP(inner_dim, config, quant_config, prefix=f"{prefix}.mlp")
def forward(
self,
@@ -181,7 +185,6 @@ class GPT2Block(nn.Module):
@support_torch_compile
class GPT2Model(nn.Module):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__()
@@ -194,20 +197,22 @@ class GPT2Model(nn.Module):
assert not config.scale_attn_by_inverse_layer_idx
assert not config.reorder_and_upcast_attn
self.embed_dim = config.hidden_size
self.wte = VocabParallelEmbedding(config.vocab_size,
self.embed_dim,
quant_config=quant_config,
prefix=f"{prefix}.wte")
self.wte = VocabParallelEmbedding(
config.vocab_size,
self.embed_dim,
quant_config=quant_config,
prefix=f"{prefix}.wte",
)
self.wpe = nn.Embedding(config.max_position_embeddings, self.embed_dim)
self.start_layer, self.end_layer, self.h = make_layers(
config.num_hidden_layers,
lambda prefix: GPT2Block(
config, cache_config, quant_config, prefix=prefix),
prefix=f"{prefix}.h")
lambda prefix: GPT2Block(config, cache_config, quant_config, prefix=prefix),
prefix=f"{prefix}.h",
)
self.ln_f = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_epsilon)
self.make_empty_intermediate_tensors = (
make_empty_intermediate_tensors_factory(["hidden_states"],
config.n_embd))
self.make_empty_intermediate_tensors = make_empty_intermediate_tensors_factory(
["hidden_states"], config.n_embd
)
def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
return self.wte(input_ids)
@@ -237,8 +242,7 @@ class GPT2Model(nn.Module):
hidden_states = self.ln_f(hidden_states)
return hidden_states
def load_weights(self, weights: Iterable[tuple[str,
torch.Tensor]]) -> set[str]:
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
params_dict = dict(self.named_parameters(remove_duplicate=False))
loaded_params: set[str] = set()
for name, loaded_weight in weights:
@@ -260,34 +264,35 @@ class GPT2Model(nn.Module):
if not name.endswith(".weight"):
continue
loaded_weight = loaded_weight.t()
weight_loader = getattr(param, "weight_loader",
default_weight_loader)
weight_loader = getattr(param, "weight_loader", default_weight_loader)
weight_loader(param, loaded_weight)
loaded_params.add(name)
return loaded_params
class GPT2LMHeadModel(nn.Module, SupportsPP):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__()
config = vllm_config.model_config.hf_config
quant_config = vllm_config.quant_config
self.config = config
self.quant_config = quant_config
self.transformer = GPT2Model(vllm_config=vllm_config,
prefix=maybe_prefix(
prefix, "transformer"))
self.lm_head = ParallelLMHead(self.config.vocab_size,
self.config.hidden_size,
quant_config=quant_config,
prefix=f"{prefix}.lm_head")
self.transformer = GPT2Model(
vllm_config=vllm_config, prefix=maybe_prefix(prefix, "transformer")
)
self.lm_head = ParallelLMHead(
self.config.vocab_size,
self.config.hidden_size,
quant_config=quant_config,
prefix=f"{prefix}.lm_head",
)
if self.config.tie_word_embeddings:
self.lm_head = self.lm_head.tie_weights(self.transformer.wte)
self.logits_processor = LogitsProcessor(config.vocab_size)
self.make_empty_intermediate_tensors = (
self.transformer.make_empty_intermediate_tensors)
self.transformer.make_empty_intermediate_tensors
)
def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
return self.transformer.get_input_embeddings(input_ids)
@@ -299,8 +304,9 @@ class GPT2LMHeadModel(nn.Module, SupportsPP):
intermediate_tensors: Optional[IntermediateTensors] = None,
inputs_embeds: Optional[torch.Tensor] = None,
) -> Union[torch.Tensor, IntermediateTensors]:
hidden_states = self.transformer(input_ids, positions,
intermediate_tensors, inputs_embeds)
hidden_states = self.transformer(
input_ids, positions, intermediate_tensors, inputs_embeds
)
return hidden_states
def compute_logits(
@@ -310,8 +316,7 @@ class GPT2LMHeadModel(nn.Module, SupportsPP):
logits = self.logits_processor(self.lm_head, hidden_states)
return logits
def load_weights(self, weights: Iterable[tuple[str,
torch.Tensor]]) -> set[str]:
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
loader = AutoWeightsLoader(self)
weights = _add_transformer_prefix(weights)
return loader.load_weights(weights)
@@ -334,22 +339,25 @@ class GPT2ForSequenceClassification(nn.Module):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__()
config = vllm_config.model_config.hf_config
self.transformer = GPT2Model(vllm_config=vllm_config,
prefix=maybe_prefix(prefix, "gpt2"))
self.score = nn.Linear(config.n_embd,
config.num_labels,
bias=False,
dtype=vllm_config.model_config.head_dtype)
self.transformer = GPT2Model(
vllm_config=vllm_config, prefix=maybe_prefix(prefix, "gpt2")
)
self.score = nn.Linear(
config.n_embd,
config.num_labels,
bias=False,
dtype=vllm_config.model_config.head_dtype,
)
pooler_config = vllm_config.model_config.pooler_config
assert pooler_config is not None
self.pooler = DispatchPooler({
"encode":
Pooler.for_encode(pooler_config),
"classify":
Pooler.for_classify(pooler_config, classifier=self.score),
})
self.pooler = DispatchPooler(
{
"encode": Pooler.for_encode(pooler_config),
"classify": Pooler.for_classify(pooler_config, classifier=self.score),
}
)
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]):
loader = AutoWeightsLoader(self)
@@ -366,15 +374,15 @@ class GPT2ForSequenceClassification(nn.Module):
input_ids=input_ids,
position_ids=positions,
inputs_embeds=inputs_embeds,
intermediate_tensors=intermediate_tensors)
intermediate_tensors=intermediate_tensors,
)
return hidden_states
def _add_transformer_prefix(
weights: Iterable[tuple[str, torch.Tensor]]
weights: Iterable[tuple[str, torch.Tensor]],
) -> Iterable[tuple[str, torch.Tensor]]:
for name, tensor in weights:
if not name.startswith('transformer.') and not name.startswith(
"lm_head"):
name = 'transformer.' + name
if not name.startswith("transformer.") and not name.startswith("lm_head"):
name = "transformer." + name
yield name, tensor