theluyuan/DiffSynth-Studio
1
0
Fork 0
mirror of https://github.com/modelscope/DiffSynth-Studio.git synced 2026-04-24 15:06:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix version issue of transformers (#1412)

Browse Source
This commit is contained in:
Zhongjie Duan
2026-04-24 17:01:08 +08:00
committed by GitHub
parent b1af4af8a9
commit 5b66f223b6
4 changed files with 15 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
diffsynth/configs/model_configs.py
View File

@@ -42,6 +42,7 @@ qwen_image_series = [
"model_hash": "5722b5c873720009de96422993b15682",
"model_name": "dinov3_image_encoder",
"model_class": "diffsynth.models.dinov3_image_encoder.DINOv3ImageEncoder",
"state_dict_converter": "diffsynth.utils.state_dict_converters.dino_v3.DINOv3StateDictConverter",
},
{
# Example:

6
diffsynth/models/dinov3_image_encoder.py
View File

@@ -1,5 +1,5 @@
from transformers import DINOv3ViTModel, DINOv3ViTImageProcessor
from transformers.models.dinov3_vit.modeling_dinov3_vit import DINOv3ViTConfig
from transformers.models.dinov3_vit.modeling_dinov3_vit import DINOv3ViTModel, DINOv3ViTConfig
from transformers import DINOv3ViTImageProcessor
import torch
from ..core.device.npu_compatible_device import get_device_type
@@ -82,7 +82,7 @@ class DINOv3ImageEncoder(DINOv3ViTModel):
hidden_states = self.embeddings(pixel_values, bool_masked_pos=bool_masked_pos)
position_embeddings = self.rope_embeddings(pixel_values)
for i, layer_module in enumerate(self.layer):
for i, layer_module in enumerate(self.model.layer):
layer_head_mask = head_mask[i] if head_mask is not None else None
hidden_states = layer_module(
hidden_states,

4
diffsynth/models/siglip2_image_encoder.py
View File

@@ -1,11 +1,11 @@
from transformers.models.siglip.modeling_siglip import SiglipVisionTransformer, SiglipVisionConfig
from transformers.models.siglip.modeling_siglip import SiglipVisionModel, SiglipVisionConfig
from transformers import SiglipImageProcessor, Siglip2VisionModel, Siglip2VisionConfig, Siglip2ImageProcessor
import torch
from diffsynth.core.device.npu_compatible_device import get_device_type
class Siglip2ImageEncoder(SiglipVisionTransformer):
class Siglip2ImageEncoder(SiglipVisionModel):
def __init__(self):
config = SiglipVisionConfig(
attention_dropout = 0.0,

9
diffsynth/utils/state_dict_converters/dino_v3.py Normal file
View File

@@ -0,0 +1,9 @@
def DINOv3StateDictConverter(state_dict):
new_state_dict = {}
for key in state_dict:
value = state_dict[key]
if key.startswith("layer"):
new_state_dict["model." + key] = value
else:
new_state_dict[key] = value
return new_state_dict