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

123
vllm/model_executor/models/olmo.py
View File

@@ -23,6 +23,7 @@
# See the License for the specific language governing permissions and
# limitations under the License.
"""Inference-only OLMo model compatible with HuggingFace weights."""
from collections.abc import Iterable
from itertools import islice
from typing import Optional, Union
@@ -36,21 +37,29 @@ 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 SiluAndMul
from vllm.model_executor.layers.linear import (MergedColumnParallelLinear,
QKVParallelLinear,
RowParallelLinear)
from vllm.model_executor.layers.linear import (
MergedColumnParallelLinear,
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 OlmoAttention(nn.Module):
@@ -70,15 +79,13 @@ class OlmoAttention(nn.Module):
super().__init__()
self.config = config
self.hidden_size = config.hidden_size
tensor_model_parallel_world_size = (
get_tensor_model_parallel_world_size())
tensor_model_parallel_world_size = get_tensor_model_parallel_world_size()
self.total_num_heads = config.num_attention_heads
assert self.hidden_size % self.total_num_heads == 0
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
self.head_dim = self.hidden_size // self.total_num_heads
self.max_position_embeddings = config.max_position_embeddings
self.rope_theta = config.rope_theta
@@ -102,12 +109,14 @@ class OlmoAttention(nn.Module):
base=self.rope_theta,
)
self.scaling = self.head_dim**-0.5
self.attn = Attention(self.num_heads,
self.head_dim,
scale=self.scaling,
cache_config=cache_config,
quant_config=quant_config,
prefix=f"{prefix}.attn")
self.attn = Attention(
self.num_heads,
self.head_dim,
scale=self.scaling,
cache_config=cache_config,
quant_config=quant_config,
prefix=f"{prefix}.attn",
)
# Attention output projection.
self.o_proj = RowParallelLinear(
@@ -189,28 +198,29 @@ class OlmoDecoderLayer(nn.Module):
(plus another skip connection).
"""
def __init__(self,
config: OlmoConfig,
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None,
prefix: str = ""):
def __init__(
self,
config: OlmoConfig,
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None,
prefix: str = "",
):
super().__init__()
# Attention block.
self.self_attn = OlmoAttention(config,
cache_config,
quant_config,
prefix=f"{prefix}.self_attn")
self.self_attn = OlmoAttention(
config, cache_config, quant_config, prefix=f"{prefix}.self_attn"
)
# MLP block.
self.mlp = OlmoMLP(config, quant_config, prefix=f"{prefix}.mlp")
# LayerNorm
self.input_layernorm = nn.LayerNorm(config.hidden_size,
elementwise_affine=False,
bias=False)
self.post_attention_layernorm = nn.LayerNorm(config.hidden_size,
elementwise_affine=False,
bias=False)
self.input_layernorm = nn.LayerNorm(
config.hidden_size, elementwise_affine=False, bias=False
)
self.post_attention_layernorm = nn.LayerNorm(
config.hidden_size, elementwise_affine=False, bias=False
)
def forward(
self,
@@ -233,7 +243,6 @@ class OlmoDecoderLayer(nn.Module):
@support_torch_compile
class OlmoModel(nn.Module):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__()
@@ -243,19 +252,22 @@ class OlmoModel(nn.Module):
self.config = 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: OlmoDecoderLayer(
config, cache_config, quant_config, prefix=prefix),
prefix=f"{prefix}.layers")
self.norm = nn.LayerNorm(config.hidden_size,
elementwise_affine=False,
bias=False)
self.make_empty_intermediate_tensors = (
make_empty_intermediate_tensors_factory(["hidden_states"],
config.hidden_size))
config, cache_config, quant_config, prefix=prefix
),
prefix=f"{prefix}.layers",
)
self.norm = nn.LayerNorm(
config.hidden_size, elementwise_affine=False, bias=False
)
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)
@@ -291,8 +303,7 @@ class OlmoModel(nn.Module):
hidden_states = self.norm(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]:
stacked_params_mapping = [
# (param_name, shard_name, shard_id)
("qkv_proj", "q_proj", "q"),
@@ -304,7 +315,7 @@ class OlmoModel(nn.Module):
params_dict = dict(self.named_parameters(remove_duplicate=False))
loaded_params: set[str] = set()
for name, loaded_weight in weights:
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)
@@ -324,8 +335,7 @@ class OlmoModel(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
@@ -335,6 +345,7 @@ class OlmoForCausalLM(nn.Module, SupportsPP, SupportsLoRA):
"""
Extremely barebones HF model wrapper.
"""
packed_modules_mapping = {
"qkv_proj": [
"q_proj",
@@ -352,8 +363,9 @@ class OlmoForCausalLM(nn.Module, SupportsPP, SupportsLoRA):
config = vllm_config.model_config.hf_config
quant_config = vllm_config.quant_config
self.config = config
self.model = OlmoModel(vllm_config=vllm_config,
prefix=maybe_prefix(prefix, "model"))
self.model = OlmoModel(
vllm_config=vllm_config, prefix=maybe_prefix(prefix, "model")
)
if config.tie_word_embeddings:
self.lm_head = self.model.embed_tokens
else:
@@ -367,7 +379,8 @@ class OlmoForCausalLM(nn.Module, SupportsPP, SupportsLoRA):
)
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)
@@ -394,11 +407,11 @@ class OlmoForCausalLM(nn.Module, SupportsPP, SupportsLoRA):
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,
skip_prefixes=(["lm_head.weight"]
if self.config.tie_word_embeddings else None),
skip_prefixes=(
["lm_head.weight"] if self.config.tie_word_embeddings else None
),
)
return loader.load_weights(weights)