[MM Encoder]: Migrate legacy ViT MultiHeadAttention to new MMEncoderAttention interface (#30684)

Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
This commit is contained in:
Isotr0py
2025-12-19 02:04:19 +08:00
committed by GitHub
parent 62be3670cb
commit 700a5ad6c6
20 changed files with 182 additions and 266 deletions
+2 -2
View File
@@ -15,7 +15,7 @@ import torch.nn as nn
import torch.nn.functional as F
from transformers import PretrainedConfig
from vllm.attention.layer import MultiHeadAttention
from vllm.attention.layers.mm_encoder_attention import MMEncoderAttention
from vllm.distributed import (
divide,
get_tensor_model_parallel_rank,
@@ -207,7 +207,7 @@ class InternParallelAttention(nn.Module):
disable_tp=use_data_parallel,
)
self.attn = MultiHeadAttention(
self.attn = MMEncoderAttention(
self.num_heads_per_partition, self.head_dim, self.scale
)