biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Model][CI] Let more pooling models support v1 (#21747)

Browse Source 
Signed-off-by: wang.yuqi <noooop@126.com>
This commit is contained in:
wang.yuqi
2025-07-31 16:51:15 +08:00
committed by GitHub
parent d2aab336ad
commit 2836dd73f1
8 changed files with 14 additions and 48 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
vllm/model_executor/models/bert_with_rope.py
View File

@@ -8,7 +8,6 @@ from torch import nn
from transformers import PretrainedConfig
from vllm.attention import Attention, AttentionType
from vllm.compilation.decorators import support_torch_compile
from vllm.config import CacheConfig, VllmConfig
from vllm.distributed import (divide, get_tensor_model_parallel_rank,
get_tensor_model_parallel_world_size,
@@ -26,7 +25,6 @@ from vllm.model_executor.layers.rotary_embedding import get_rope
from vllm.model_executor.layers.vocab_parallel_embedding import (
VocabParallelEmbedding)
from vllm.model_executor.model_loader.weight_utils import default_weight_loader
from vllm.model_executor.models import SupportsV0Only
from vllm.model_executor.models.interfaces import SupportsQuant
from vllm.model_executor.models.utils import WeightsMapper
from vllm.model_executor.utils import set_weight_attrs
@@ -360,7 +358,6 @@ class BertWithRopeBlock(nn.Module):
return hidden_states
@support_torch_compile
class BertWithRopeEncoder(nn.Module):
def __init__(self,
@@ -394,7 +391,7 @@ class BertWithRopeEncoder(nn.Module):
return hidden_states
class BertWithRope(nn.Module, SupportsV0Only, SupportsQuant):
class BertWithRope(nn.Module, SupportsQuant):
hf_to_vllm_mapper = WeightsMapper(orig_to_new_prefix={"model.": ""})
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):

2
vllm/model_executor/models/config.py
View File

@@ -93,7 +93,7 @@ class NomicBertModelConfig(VerifyAndUpdateConfig):
config.num_hidden_layers = config.n_layer
head_dim = config.hidden_size // config.num_attention_heads
rotary_emb_dim = head_dim * config.rotary_emb_fraction
rotary_emb_dim = int(head_dim * config.rotary_emb_fraction)
max_trained_positions = getattr(config, "max_trained_positions", 2048)
config.rotary_kwargs = {
"head_size": head_dim,

2
vllm/model_executor/models/modernbert.py
View File

@@ -8,7 +8,6 @@ from torch import nn
from transformers import ModernBertConfig
from vllm.attention import Attention, AttentionType
from vllm.compilation.decorators import support_torch_compile
from vllm.config import VllmConfig
from vllm.distributed import get_tensor_model_parallel_world_size
from vllm.model_executor.layers.linear import (QKVParallelLinear,
@@ -200,7 +199,6 @@ class ModernBertEncoderLayer(nn.Module):
return hidden_states
@support_torch_compile
class ModernBertModel(nn.Module):
hf_to_vllm_mapper = WeightsMapper(
orig_to_new_prefix={"layers.": "encoder_layer.layers."})