[Diffusion][Attention] Support per-role attention backend via CLI (#2681)

Signed-off-by: gcanlin <canlinguosdu@gmail.com>
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>
This commit is contained in:
Canlin GuoandHongsheng Liu authored and GitHub committed 2026-05-10 11:59:05 +08:00
1 parent 26d481fc84
commit 11c4c7f0ff
29 files changed
+1262 -106

No files matched your search

@@ -32,6 +32,7 @@ import pytest
import torch
from vllm_omni.diffusion.attention.layer import Attention
from vllm_omni.diffusion.config import set_current_diffusion_config
from vllm_omni.diffusion.data import (
DiffusionParallelConfig,
OmniDiffusionConfig,
@@ -403,8 +404,8 @@ def ulysses_attention_on_test_model(
pipeline_parallel_size=1,
)
# Set the config so Attention can access it
with set_forward_context(omni_diffusion_config=od_config):
# Set the config so Attention can access it during init and forward
with set_forward_context(omni_diffusion_config=od_config), set_current_diffusion_config(od_config):
# Create model
hidden_size = num_heads * head_size