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

116
vllm/model_executor/models/gemma.py
View File

@@ -16,6 +16,7 @@
# See the License for the specific language governing permissions and
# limitations under the License.
"""Inference-only Gemma model compatible with HuggingFace weights."""
from collections.abc import Iterable
from functools import cache
from itertools import islice
@@ -32,21 +33,26 @@ from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
from vllm.logger import init_logger
from vllm.model_executor.layers.activation import GeluAndMul
from vllm.model_executor.layers.layernorm import GemmaRMSNorm
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 (
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 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,
)
logger = init_logger(__name__)
@@ -66,19 +72,22 @@ def _get_gemma_act_fn(
"`%s`, edit the config JSON to set "
"`hidden_activation=%s` instead of `hidden_act`. "
"See https://github.com/huggingface/transformers/pull/29402 "
"for more details.", hidden_act, hidden_act)
"for more details.",
hidden_act,
hidden_act,
)
return GeluAndMul(approximate="tanh")
elif hidden_activation == "gelu_pytorch_tanh":
return GeluAndMul(approximate="tanh")
elif hidden_activation == "gelu":
return GeluAndMul(approximate="none")
else:
raise ValueError(f"Activation function {hidden_act} is not "
"supported for Gemma models.")
raise ValueError(
f"Activation function {hidden_act} is not supported for Gemma models."
)
class GemmaMLP(nn.Module):
def __init__(
self,
hidden_size: int,
@@ -113,7 +122,6 @@ class GemmaMLP(nn.Module):
class GemmaAttention(nn.Module):
def __init__(
self,
hidden_size: int,
@@ -172,13 +180,15 @@ class GemmaAttention(nn.Module):
base=self.rope_theta,
is_neox_style=True,
)
self.attn = Attention(self.num_heads,
self.head_dim,
self.scaling,
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,
self.scaling,
num_kv_heads=self.num_kv_heads,
cache_config=cache_config,
quant_config=quant_config,
prefix=f"{prefix}.attn",
)
def forward(
self,
@@ -194,7 +204,6 @@ class GemmaAttention(nn.Module):
class GemmaDecoderLayer(nn.Module):
def __init__(
self,
config: GemmaConfig,
@@ -223,10 +232,10 @@ class GemmaDecoderLayer(nn.Module):
quant_config=quant_config,
prefix=f"{prefix}.mlp",
)
self.input_layernorm = GemmaRMSNorm(config.hidden_size,
eps=config.rms_norm_eps)
self.post_attention_layernorm = GemmaRMSNorm(config.hidden_size,
eps=config.rms_norm_eps)
self.input_layernorm = GemmaRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.post_attention_layernorm = GemmaRMSNorm(
config.hidden_size, eps=config.rms_norm_eps
)
def forward(
self,
@@ -239,23 +248,20 @@ class GemmaDecoderLayer(nn.Module):
residual = hidden_states
hidden_states = self.input_layernorm(hidden_states)
else:
hidden_states, residual = self.input_layernorm(
hidden_states, residual)
hidden_states, residual = self.input_layernorm(hidden_states, residual)
hidden_states = self.self_attn(
positions=positions,
hidden_states=hidden_states,
)
# Fully Connected
hidden_states, residual = self.post_attention_layernorm(
hidden_states, residual)
hidden_states, residual = self.post_attention_layernorm(hidden_states, residual)
hidden_states = self.mlp(hidden_states)
return hidden_states, residual
@support_torch_compile
class GemmaModel(nn.Module):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__()
@@ -272,8 +278,10 @@ class GemmaModel(nn.Module):
self.start_layer, self.end_layer, self.layers = make_layers(
config.num_hidden_layers,
lambda prefix: GemmaDecoderLayer(
config, cache_config, quant_config, prefix=prefix),
prefix=f"{prefix}.layers")
config, cache_config, quant_config, prefix=prefix
),
prefix=f"{prefix}.layers",
)
self.norm = GemmaRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
# Normalize the embedding by sqrt(hidden_size)
@@ -281,12 +289,10 @@ class GemmaModel(nn.Module):
# data type such as bfloat16, not float32.
# See https://github.com/huggingface/transformers/pull/29402
normalizer = self.config.hidden_size**0.5
self.register_buffer("normalizer",
torch.tensor(normalizer),
persistent=False)
self.make_empty_intermediate_tensors = (
make_empty_intermediate_tensors_factory(
["hidden_states", "residual"], config.hidden_size))
self.register_buffer("normalizer", torch.tensor(normalizer), persistent=False)
self.make_empty_intermediate_tensors = make_empty_intermediate_tensors_factory(
["hidden_states", "residual"], config.hidden_size
)
def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
return self.embed_tokens(input_ids)
@@ -315,15 +321,13 @@ class GemmaModel(nn.Module):
residual,
)
if not get_pp_group().is_last_rank:
return IntermediateTensors({
"hidden_states": hidden_states,
"residual": residual
})
return IntermediateTensors(
{"hidden_states": hidden_states, "residual": residual}
)
hidden_states, _ = self.norm(hidden_states, residual)
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"),
@@ -335,7 +339,7 @@ class GemmaModel(nn.Module):
params_dict = dict(self.named_parameters())
loaded_params: set[str] = set()
for name, loaded_weight in weights:
for (param_name, shard_name, shard_id) in stacked_params_mapping:
for param_name, shard_name, shard_id in stacked_params_mapping:
if shard_name not in name:
continue
name = name.replace(shard_name, param_name)
@@ -355,8 +359,7 @@ class GemmaModel(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)
@@ -388,11 +391,13 @@ class GemmaForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
self.lora_config = lora_config
self.quant_config = quant_config
self.model = GemmaModel(vllm_config=vllm_config,
prefix=maybe_prefix(prefix, "model"))
self.model = GemmaModel(
vllm_config=vllm_config, prefix=maybe_prefix(prefix, "model")
)
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)
@@ -404,8 +409,9 @@ class GemmaForCausalLM(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
def compute_logits(
@@ -415,11 +421,9 @@ class GemmaForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
logits = self.logits_processor(self.model.embed_tokens, 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."]
if self.config.tie_word_embeddings else None),
skip_prefixes=(["lm_head."] if self.config.tie_word_embeddings else None),
)
return loader.load_weights(weights)