[Diffusion] switch to local calculate_dimensions (#13294)

python/sglang/multimodal_gen/configs/models/vaes/qwenimage.py

@@ -3,9 +3,8 @@
 # SPDX-License-Identifier: Apache-2.0
 from dataclasses import dataclass, field
 
-from diffusers.pipelines.qwenimage.pipeline_qwenimage_edit import calculate_dimensions
-
 from sglang.multimodal_gen.configs.models.vaes.base import VAEArchConfig, VAEConfig
+from sglang.multimodal_gen.utils import calculate_dimensions
 
 
 @dataclass

python/sglang/multimodal_gen/configs/pipelines/qwen_image.py

@@ -4,13 +4,13 @@ from dataclasses import dataclass, field
 from typing import Callable
 
 import torch
-from diffusers.pipelines.qwenimage.pipeline_qwenimage_edit import calculate_dimensions
 
 from sglang.multimodal_gen.configs.models import DiTConfig, EncoderConfig, VAEConfig
 from sglang.multimodal_gen.configs.models.dits.qwenimage import QwenImageDitConfig
 from sglang.multimodal_gen.configs.models.encoders.qwen_image import Qwen2_5VLConfig
 from sglang.multimodal_gen.configs.models.vaes.qwenimage import QwenImageVAEConfig
 from sglang.multimodal_gen.configs.pipelines.base import ModelTaskType, PipelineConfig
+from sglang.multimodal_gen.utils import calculate_dimensions
 
 
 def _extract_masked_hidden(hidden_states: torch.Tensor, mask: torch.Tensor):

python/sglang/multimodal_gen/utils.py

@@ -801,3 +801,13 @@ def save_decoded_latents_as_video(
 
     os.makedirs(os.path.dirname(output_path), exist_ok=True)
     imageio.mimsave(output_path, frames, fps=fps, format="mp4")
+
+
+def calculate_dimensions(target_area, ratio):
+    width = math.sqrt(target_area * ratio)
+    height = width / ratio
+
+    width = round(width / 32) * 32
+    height = round(height / 32) * 32
+
+    return width, height, None