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

174
vllm/model_executor/models/qwen2.py
View File

@@ -24,6 +24,7 @@
# See the License for the specific language governing permissions and
# limitations under the License.
"""Inference-only Qwen2 model compatible with HuggingFace weights."""
from collections.abc import Iterable
from itertools import islice
from typing import Any, Optional, Union
@@ -39,28 +40,38 @@ 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.layernorm import RMSNorm
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, maybe_remap_kv_scale_name)
default_weight_loader,
maybe_remap_kv_scale_name,
)
from vllm.sequence import IntermediateTensors
from vllm.transformers_utils.config import is_interleaved
from .interfaces import SupportsEagle3, SupportsLoRA, SupportsPP
from .utils import (AutoWeightsLoader, PPMissingLayer, extract_layer_index,
is_pp_missing_parameter,
make_empty_intermediate_tensors_factory, make_layers,
maybe_prefix)
from .utils import (
AutoWeightsLoader,
PPMissingLayer,
extract_layer_index,
is_pp_missing_parameter,
make_empty_intermediate_tensors_factory,
make_layers,
maybe_prefix,
)
class Qwen2MLP(nn.Module):
def __init__(
self,
hidden_size: int,
@@ -85,8 +96,9 @@ class Qwen2MLP(nn.Module):
prefix=f"{prefix}.down_proj",
)
if hidden_act != "silu":
raise ValueError(f"Unsupported activation: {hidden_act}. "
"Only silu is supported for now.")
raise ValueError(
f"Unsupported activation: {hidden_act}. Only silu is supported for now."
)
self.act_fn = SiluAndMul()
def forward(self, x):
@@ -97,7 +109,6 @@ class Qwen2MLP(nn.Module):
class Qwen2Attention(nn.Module):
def __init__(
self,
hidden_size: int,
@@ -160,8 +171,11 @@ class Qwen2Attention(nn.Module):
rope_scaling=rope_scaling,
dual_chunk_attention_config=dual_chunk_attention_config,
)
attn_cls = (EncoderOnlyAttention
if attn_type == AttentionType.ENCODER_ONLY else Attention)
attn_cls = (
EncoderOnlyAttention
if attn_type == AttentionType.ENCODER_ONLY
else Attention
)
self.attn = attn_cls(
self.num_heads,
self.head_dim,
@@ -174,7 +188,10 @@ class Qwen2Attention(nn.Module):
**{
"layer_idx": extract_layer_index(prefix),
"dual_chunk_attention_config": dual_chunk_attention_config,
} if dual_chunk_attention_config else {})
}
if dual_chunk_attention_config
else {},
)
def forward(
self,
@@ -190,7 +207,6 @@ class Qwen2Attention(nn.Module):
class Qwen2DecoderLayer(nn.Module):
def __init__(
self,
config: Qwen2Config,
@@ -203,9 +219,9 @@ class Qwen2DecoderLayer(nn.Module):
# Requires transformers > 4.32.0
rope_theta = getattr(config, "rope_theta", 1000000)
rope_scaling = getattr(config, "rope_scaling", None)
dual_chunk_attention_config = getattr(config,
"dual_chunk_attention_config",
None)
dual_chunk_attention_config = getattr(
config, "dual_chunk_attention_config", None
)
# By default, Qwen2 uses causal attention as it is a decoder-only model.
# You can override the HF config with `is_causal=False` to enable
@@ -236,10 +252,10 @@ class Qwen2DecoderLayer(nn.Module):
quant_config=quant_config,
prefix=f"{prefix}.mlp",
)
self.input_layernorm = RMSNorm(config.hidden_size,
eps=config.rms_norm_eps)
self.post_attention_layernorm = RMSNorm(config.hidden_size,
eps=config.rms_norm_eps)
self.input_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.post_attention_layernorm = RMSNorm(
config.hidden_size, eps=config.rms_norm_eps
)
def forward(
self,
@@ -252,16 +268,14 @@ class Qwen2DecoderLayer(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
@@ -274,14 +288,16 @@ class Qwen2DecoderLayer(nn.Module):
"positions": -1,
"intermediate_tensors": 0,
"inputs_embeds": 0,
})
}
)
class Qwen2Model(nn.Module):
def __init__(self,
*,
vllm_config: VllmConfig,
prefix: str = "",
decoder_layer_type: type[nn.Module] = Qwen2DecoderLayer):
def __init__(
self,
*,
vllm_config: VllmConfig,
prefix: str = "",
decoder_layer_type: type[nn.Module] = Qwen2DecoderLayer,
):
super().__init__()
config = vllm_config.model_config.hf_config.get_text_config()
@@ -297,14 +313,16 @@ class Qwen2Model(nn.Module):
"to discuss this feature.".format(
config.max_window_layers,
config.num_hidden_layers,
))
)
)
self.config = config
self.quant_config = quant_config
self.vocab_size = config.vocab_size
if get_pp_group().is_first_rank or (config.tie_word_embeddings
and get_pp_group().is_last_rank):
if get_pp_group().is_first_rank or (
config.tie_word_embeddings and get_pp_group().is_last_rank
):
self.embed_tokens = VocabParallelEmbedding(
config.vocab_size,
config.hidden_size,
@@ -318,16 +336,18 @@ class Qwen2Model(nn.Module):
decoder_layer_type = decoder_layer_type or Qwen2DecoderLayer
self.start_layer, self.end_layer, self.layers = make_layers(
config.num_hidden_layers,
lambda prefix: decoder_layer_type(config=config,
cache_config=cache_config,
quant_config=quant_config,
prefix=prefix),
lambda prefix: decoder_layer_type(
config=config,
cache_config=cache_config,
quant_config=quant_config,
prefix=prefix,
),
prefix=f"{prefix}.layers",
)
self.make_empty_intermediate_tensors = (
make_empty_intermediate_tensors_factory(
["hidden_states", "residual"], config.hidden_size))
self.make_empty_intermediate_tensors = make_empty_intermediate_tensors_factory(
["hidden_states", "residual"], config.hidden_size
)
if get_pp_group().is_last_rank:
self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
else:
@@ -358,16 +378,16 @@ class Qwen2Model(nn.Module):
aux_hidden_states = []
for idx, layer in enumerate(
islice(self.layers, self.start_layer, self.end_layer)):
islice(self.layers, self.start_layer, self.end_layer)
):
if idx in self.aux_hidden_state_layers:
aux_hidden_states.append(hidden_states + residual)
hidden_states, residual = layer(positions, hidden_states, 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)
@@ -376,8 +396,7 @@ class Qwen2Model(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"),
@@ -391,18 +410,19 @@ class Qwen2Model(nn.Module):
for name, loaded_weight in weights:
if "rotary_emb.inv_freq" in name:
continue
if (self.quant_config is not None and
(scale_name := self.quant_config.get_cache_scale(name))):
if self.quant_config is not None and (
scale_name := self.quant_config.get_cache_scale(name)
):
# Loading kv cache quantization scales
param = params_dict[scale_name]
weight_loader = getattr(param, "weight_loader",
default_weight_loader)
loaded_weight = (loaded_weight if loaded_weight.dim() == 0 else
loaded_weight[0])
weight_loader = getattr(param, "weight_loader", default_weight_loader)
loaded_weight = (
loaded_weight if loaded_weight.dim() == 0 else loaded_weight[0]
)
weight_loader(param, loaded_weight)
loaded_params.add(scale_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)
@@ -417,8 +437,7 @@ class Qwen2Model(nn.Module):
if name is None:
continue
param = params_dict[name]
weight_loader = getattr(param, "weight_loader",
default_weight_loader)
weight_loader = getattr(param, "weight_loader", default_weight_loader)
if weight_loader == default_weight_loader:
weight_loader(param, loaded_weight)
else:
@@ -435,8 +454,7 @@ class Qwen2Model(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
@@ -465,25 +483,28 @@ class Qwen2ForCausalLM(nn.Module, SupportsLoRA, SupportsPP, SupportsEagle3):
self.lora_config = lora_config
self.quant_config = quant_config
self.model = Qwen2Model(vllm_config=vllm_config,
prefix=maybe_prefix(prefix, "model"))
self.model = Qwen2Model(
vllm_config=vllm_config, prefix=maybe_prefix(prefix, "model")
)
if get_pp_group().is_last_rank:
if config.tie_word_embeddings:
self.lm_head = self.model.embed_tokens
else:
self.lm_head = ParallelLMHead(config.vocab_size,
config.hidden_size,
quant_config=quant_config,
prefix=maybe_prefix(
prefix, "lm_head"))
self.lm_head = ParallelLMHead(
config.vocab_size,
config.hidden_size,
quant_config=quant_config,
prefix=maybe_prefix(prefix, "lm_head"),
)
else:
self.lm_head = PPMissingLayer()
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)
@@ -502,8 +523,9 @@ class Qwen2ForCausalLM(nn.Module, SupportsLoRA, SupportsPP, SupportsEagle3):
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(
@@ -513,11 +535,9 @@ class Qwen2ForCausalLM(nn.Module, SupportsLoRA, SupportsPP, SupportsEagle3):
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."]
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)