[New Model]: google/embeddinggemma-300m (#24318)

Signed-off-by: wang.yuqi <noooop@126.com>
This commit is contained in:
wang.yuqi
2025-09-05 22:58:36 -07:00
committed by GitHub
parent 53b19ccdd5
commit 6d6c6b05d3
9 changed files with 73 additions and 29 deletions
+20 -11
View File
@@ -24,7 +24,7 @@ import torch.nn.functional as F
from torch import nn
from transformers import Gemma3TextConfig
from vllm.attention import Attention
from vllm.attention import Attention, AttentionType
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
@@ -44,6 +44,7 @@ from vllm.model_executor.model_loader.weight_utils import (
from vllm.model_executor.sampling_metadata import SamplingMetadata
from vllm.sequence import IntermediateTensors
from ...attention.layers.encoder_only_attention import EncoderOnlyAttention
from .interfaces import SupportsLoRA, SupportsPP
from .utils import (AutoWeightsLoader, extract_layer_index,
is_pp_missing_parameter,
@@ -169,16 +170,24 @@ class Gemma3Attention(nn.Module):
rope_scaling=self.rope_scaling,
)
# Initialize the attention.
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,
logits_soft_cap=attn_logits_soft_cap,
per_layer_sliding_window=sliding_window,
prefix=f"{prefix}.attn")
if getattr(config, "is_causal", True):
attn_type = AttentionType.DECODER
else:
attn_type = AttentionType.ENCODER_ONLY
attn_cls = (EncoderOnlyAttention
if attn_type == AttentionType.ENCODER_ONLY else Attention)
self.attn = attn_cls(self.num_heads,
self.head_dim,
self.scaling,
num_kv_heads=self.num_kv_heads,
cache_config=cache_config,
quant_config=quant_config,
attn_type=attn_type,
logits_soft_cap=attn_logits_soft_cap,
per_layer_sliding_window=sliding_window,
prefix=f"{prefix}.attn")
def forward(
self,