Merge branch 'huggingface:main' into add-neuron-backend

Author: JingyaHuang

src/diffusers/models/transformers/transformer_helios.py

@@ -13,7 +13,6 @@
 # limitations under the License.
 
 import math
-from functools import lru_cache
 from typing import Any
 
 import torch
@@ -343,7 +342,6 @@ def get_frequency_batched(self, freqs_base, pos):
         return freqs.cos(), freqs.sin()
 
     @torch.no_grad()
-    @lru_cache(maxsize=32)
     def _get_spatial_meshgrid(self, height, width, device_str):
         device = torch.device(device_str)
         grid_y_coords = torch.arange(height, device=device, dtype=torch.float32)

src/diffusers/pipelines/latent_diffusion/pipeline_latent_diffusion.py

@@ -720,6 +720,7 @@ def __init__(self, config: LDMBertConfig):
         super().__init__(config)
         self.model = LDMBertEncoder(config)
         self.to_logits = nn.Linear(config.hidden_size, config.vocab_size)
+        self.post_init()
 
     def forward(
         self,

src/diffusers/pipelines/paint_by_example/image_encoder.py

@@ -35,6 +35,8 @@ def __init__(self, config, proj_size=None):
         # uncondition for scaling
         self.uncond_vector = nn.Parameter(torch.randn((1, 1, self.proj_size)))
 
+        self.post_init()
+
     def forward(self, pixel_values, return_uncond_vector=False):
         clip_output = self.model(pixel_values=pixel_values)
         latent_states = clip_output.pooler_output