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

100
vllm/model_executor/models/mpt.py
View File

@@ -14,30 +14,38 @@ from transformers import MptConfig
from vllm.attention import Attention
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_rank,
get_tensor_model_parallel_world_size)
from vllm.distributed import (
get_pp_group,
get_tensor_model_parallel_rank,
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 (
VocabParallelEmbedding)
from vllm.model_executor.layers.vocab_parallel_embedding import VocabParallelEmbedding
from vllm.model_executor.model_loader.weight_utils import default_weight_loader
from vllm.sequence import IntermediateTensors
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,
)
def _get_alibi_slopes(
total_num_heads: int,
alibi_bias_max: int,
) -> torch.Tensor:
next_power_of_2 = 2**math.ceil(math.log2(total_num_heads))
next_power_of_2 = 2 ** math.ceil(math.log2(total_num_heads))
m = torch.arange(1, next_power_of_2 + 1, dtype=torch.float32)
m = m.mul(alibi_bias_max / next_power_of_2)
slopes = 1.0 / torch.pow(2, m)
@@ -47,7 +55,6 @@ def _get_alibi_slopes(
class MPTAttention(nn.Module):
def __init__(
self,
config: MptConfig,
@@ -107,20 +114,21 @@ class MPTAttention(nn.Module):
tp_rank = get_tensor_model_parallel_rank()
head_start = tp_rank * self.num_heads
head_end = (tp_rank + 1) * self.num_heads
alibi_slopes = _get_alibi_slopes(self.total_num_heads,
self.alibi_bias_max)
alibi_slopes = _get_alibi_slopes(self.total_num_heads, self.alibi_bias_max)
alibi_slopes = alibi_slopes[head_start:head_end].tolist()
self.head_dim = self.d_model // self.total_num_heads
scaling = self.head_dim**-0.5
self.attn = Attention(self.num_heads,
self.head_dim,
scaling,
alibi_slopes=alibi_slopes,
num_kv_heads=self.num_kv_heads,
cache_config=cache_config,
quant_config=quant_config,
prefix=f"{prefix}.attn")
self.attn = Attention(
self.num_heads,
self.head_dim,
scaling,
alibi_slopes=alibi_slopes,
num_kv_heads=self.num_kv_heads,
cache_config=cache_config,
quant_config=quant_config,
prefix=f"{prefix}.attn",
)
def forward(
self,
@@ -141,7 +149,6 @@ class MPTAttention(nn.Module):
class MPTMLP(nn.Module):
def __init__(
self,
config: MptConfig,
@@ -173,7 +180,6 @@ class MPTMLP(nn.Module):
class MPTBlock(nn.Module):
def __init__(
self,
config: MptConfig,
@@ -184,10 +190,9 @@ class MPTBlock(nn.Module):
super().__init__()
hidden_size = config.d_model
self.norm_1 = nn.LayerNorm(hidden_size)
self.attn = MPTAttention(config,
cache_config,
quant_config,
prefix=f"{prefix}.attn")
self.attn = MPTAttention(
config, cache_config, quant_config, prefix=f"{prefix}.attn"
)
self.norm_2 = nn.LayerNorm(hidden_size)
self.ffn = MPTMLP(config, quant_config)
@@ -210,7 +215,6 @@ class MPTBlock(nn.Module):
@support_torch_compile
class MPTModel(nn.Module):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__()
@@ -227,19 +231,18 @@ class MPTModel(nn.Module):
)
self.start_layer, self.end_layer, self.blocks = make_layers(
config.n_layers,
lambda prefix: MPTBlock(
config, cache_config, quant_config, prefix=prefix),
prefix=f"{prefix}.blocks")
lambda prefix: MPTBlock(config, cache_config, quant_config, prefix=prefix),
prefix=f"{prefix}.blocks",
)
self.norm_f = nn.LayerNorm(config.d_model)
if config.no_bias:
for module in self.modules():
if hasattr(module, "bias") and isinstance(
module.bias, nn.Parameter):
if hasattr(module, "bias") and isinstance(module.bias, nn.Parameter):
# Remove the bias term in Linear and LayerNorm.
module.register_parameter("bias", None)
self.make_empty_intermediate_tensors = (
make_empty_intermediate_tensors_factory(["hidden_states"],
config.d_model))
self.make_empty_intermediate_tensors = make_empty_intermediate_tensors_factory(
["hidden_states"], config.d_model
)
def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
return self.wte(input_ids)
@@ -267,8 +270,7 @@ class MPTModel(nn.Module):
hidden_states = self.norm_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:
@@ -278,15 +280,13 @@ class MPTModel(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
class MPTForCausalLM(nn.Module, SupportsPP):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__()
config = vllm_config.model_config.hf_config
@@ -295,12 +295,14 @@ class MPTForCausalLM(nn.Module, SupportsPP):
assert config.tie_word_embeddings
self.quant_config = quant_config
self.transformer = MPTModel(vllm_config=vllm_config,
prefix=maybe_prefix(prefix, "transformer"))
self.transformer = MPTModel(
vllm_config=vllm_config, prefix=maybe_prefix(prefix, "transformer")
)
self.lm_head = 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)
@@ -312,8 +314,9 @@ class MPTForCausalLM(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(
@@ -323,7 +326,6 @@ class MPTForCausalLM(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)
return loader.load_weights(weights)