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

161
vllm/model_executor/models/phi.py
View File

@@ -37,6 +37,7 @@
# OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
"""Inference-only Phi-1.5 model compatible with HuggingFace weights."""
from collections.abc import Iterable
from itertools import islice
from typing import Optional, Union
@@ -50,40 +51,47 @@ from vllm.compilation.decorators import support_torch_compile
from vllm.config import CacheConfig, VllmConfig
from vllm.distributed import 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.rotary_embedding import get_rope
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 .interfaces import SupportsLoRA, 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 PhiAttention(nn.Module):
def __init__(self,
config: PhiConfig,
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None,
prefix: str = ""):
def __init__(
self,
config: PhiConfig,
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None,
prefix: str = "",
):
super().__init__()
self.total_num_heads = config.num_attention_heads
self.hidden_size = config.hidden_size
self.head_size = self.hidden_size // self.total_num_heads
tensor_model_parallel_world_size = (
get_tensor_model_parallel_world_size())
tensor_model_parallel_world_size = get_tensor_model_parallel_world_size()
assert self.total_num_heads % tensor_model_parallel_world_size == 0
self.num_heads = (self.total_num_heads //
tensor_model_parallel_world_size)
self.num_heads = self.total_num_heads // tensor_model_parallel_world_size
# pylint: disable=C0103
self.qkv_proj = QKVParallelLinear(
@@ -100,28 +108,31 @@ class PhiAttention(nn.Module):
)
scaling = self.head_size**-0.5
rotary_dim = int(config.partial_rotary_factor *
(config.hidden_size // config.num_attention_heads))
rotary_dim = int(
config.partial_rotary_factor
* (config.hidden_size // config.num_attention_heads)
)
assert rotary_dim % 2 == 0
# pylint: disable=C0301
# Refer to:
# https://huggingface.co/microsoft/phi-1_5/blob/d212a789620c380ff32ca1d1ee9943a777360987/modeling_phi.py#L518
rope_theta = getattr(config, "rope_theta", 10000.0)
max_position_embeddings = getattr(config, "max_position_embeddings",
2048)
max_position_embeddings = getattr(config, "max_position_embeddings", 2048)
self.rotary_emb = get_rope(
self.head_size,
rotary_dim=rotary_dim,
max_position=max_position_embeddings,
base=rope_theta,
)
self.attn = Attention(self.num_heads,
self.head_size,
scaling,
cache_config=cache_config,
quant_config=quant_config,
prefix=f"{prefix}.attn")
self.attn = Attention(
self.num_heads,
self.head_size,
scaling,
cache_config=cache_config,
quant_config=quant_config,
prefix=f"{prefix}.attn",
)
def forward(
self,
@@ -137,10 +148,9 @@ class PhiAttention(nn.Module):
class PhiMLP(nn.Module):
def __init__(self,
config: PhiConfig,
quant_config: Optional[QuantizationConfig] = None):
def __init__(
self, config: PhiConfig, quant_config: Optional[QuantizationConfig] = None
):
super().__init__()
n_inner = getattr(config, "n_inner", None)
@@ -166,19 +176,20 @@ class PhiMLP(nn.Module):
class PhiLayer(nn.Module):
def __init__(self,
config: PhiConfig,
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None,
prefix: str = ""):
def __init__(
self,
config: PhiConfig,
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None,
prefix: str = "",
):
super().__init__()
self.input_layernorm = nn.LayerNorm(config.hidden_size,
eps=config.layer_norm_eps)
self.self_attn = PhiAttention(config,
cache_config,
quant_config,
prefix=f"{prefix}.self_attn")
self.input_layernorm = nn.LayerNorm(
config.hidden_size, eps=config.layer_norm_eps
)
self.self_attn = PhiAttention(
config, cache_config, quant_config, prefix=f"{prefix}.self_attn"
)
self.mlp = PhiMLP(config, quant_config)
def forward(
@@ -199,7 +210,6 @@ class PhiLayer(nn.Module):
@support_torch_compile
class PhiModel(nn.Module):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__()
@@ -209,18 +219,20 @@ class PhiModel(nn.Module):
self.config = config
self.quant_config = quant_config
self.embed_tokens = VocabParallelEmbedding(config.vocab_size,
config.hidden_size)
self.embed_tokens = VocabParallelEmbedding(
config.vocab_size, config.hidden_size
)
self.start_layer, self.end_layer, self.layers = make_layers(
config.num_hidden_layers,
lambda prefix: PhiLayer(
config, cache_config, quant_config, prefix=prefix),
prefix=f"{prefix}.layers")
self.final_layernorm = nn.LayerNorm(config.hidden_size,
eps=config.layer_norm_eps)
self.make_empty_intermediate_tensors = (
make_empty_intermediate_tensors_factory(["hidden_states"],
config.hidden_size))
lambda prefix: PhiLayer(config, cache_config, quant_config, prefix=prefix),
prefix=f"{prefix}.layers",
)
self.final_layernorm = nn.LayerNorm(
config.hidden_size, eps=config.layer_norm_eps
)
self.make_empty_intermediate_tensors = make_empty_intermediate_tensors_factory(
["hidden_states"], config.hidden_size
)
def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
return self.embed_tokens(input_ids)
@@ -250,13 +262,12 @@ class PhiModel(nn.Module):
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]:
stacked_params_mapping = [
# (param_name, shard_name, shard_id)
("qkv_proj", "q_proj", "q"),
("qkv_proj", "k_proj", "k"),
("qkv_proj", "v_proj", "v")
("qkv_proj", "v_proj", "v"),
]
params_dict = dict(self.named_parameters())
loaded_params: set[str] = set()
@@ -265,7 +276,7 @@ class PhiModel(nn.Module):
if "rotary_emb.inv_freq" in name:
continue
for (param_name, weight_name, shard_id) in stacked_params_mapping:
for param_name, weight_name, shard_id in stacked_params_mapping:
if weight_name not in name:
continue
name = name.replace(weight_name, param_name)
@@ -287,8 +298,7 @@ class PhiModel(nn.Module):
if is_pp_missing_parameter(name, self):
continue
param = params_dict[name]
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
@@ -315,17 +325,21 @@ class PhiForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
self.quant_config = quant_config
self.model = PhiModel(vllm_config=vllm_config,
prefix=maybe_prefix(prefix, "model"))
self.model = PhiModel(
vllm_config=vllm_config, prefix=maybe_prefix(prefix, "model")
)
self.lm_head = ParallelLMHead(config.vocab_size,
config.hidden_size,
bias=True,
quant_config=quant_config,
prefix=maybe_prefix(prefix, "lm_head"))
self.lm_head = ParallelLMHead(
config.vocab_size,
config.hidden_size,
bias=True,
quant_config=quant_config,
prefix=maybe_prefix(prefix, "lm_head"),
)
self.logits_processor = LogitsProcessor(config.vocab_size)
self.make_empty_intermediate_tensors = (
self.model.make_empty_intermediate_tensors)
self.model.make_empty_intermediate_tensors
)
def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
return self.model.get_input_embeddings(input_ids)
@@ -337,8 +351,9 @@ class PhiForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
intermediate_tensors: Optional[IntermediateTensors] = None,
inputs_embeds: Optional[torch.Tensor] = None,
) -> Union[torch.Tensor, IntermediateTensors]:
hidden_states = self.model(input_ids, positions, intermediate_tensors,
inputs_embeds)
hidden_states = self.model(
input_ids, positions, intermediate_tensors, inputs_embeds
)
return hidden_states
@@ -346,11 +361,9 @@ class PhiForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
self,
hidden_states: torch.Tensor,
) -> Optional[torch.Tensor]:
logits = self.logits_processor(self.lm_head, hidden_states,
self.lm_head.bias)
logits = self.logits_processor(self.lm_head, hidden_states, self.lm_head.bias)
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)
return loader.load_weights(weights)