Anyone had this error while trying to load model into pipeline?
I got this error and was stuck for several days when the model was loaded into memory. I have tried re-downloading the model files and still got this error. Any thoughts or solutions, thanks.
ValueError: Cannot load <class 'Pyramid_Flow.pyramid_dit.mmdit_modules.modeling_pyramid_mmdit.PyramidDiffusionMMDiT'> from Pyramid_Flow/Pyramid-Flow-Miniflux/diffusion_transformer_384p because the following keys are missing:
transformer_blocks.3.attn.norm_add_k.weight, transformer_blocks.5.attn.norm_add_k.weight, pos_embed.proj.weight, transformer_blocks.5.attn.norm_add_q.weight, transformer_blocks.7.attn.norm_add_q.weight, transformer_blocks.0.attn.norm_add_q.weight, transformer_blocks.6.attn.norm_add_q.weight, pos_embed.proj.bias, transformer_blocks.0.attn.norm_add_k.weight, transformer_blocks.1.attn.norm_add_k.weight, transformer_blocks.1.attn.norm_add_q.weight, transformer_blocks.3.attn.norm_add_q.weight, transformer_blocks.2.attn.norm_add_q.weight, transformer_blocks.6.attn.norm_add_k.weight, transformer_blocks.4.attn.norm_add_k.weight, transformer_blocks.7.attn.norm_add_k.weight, transformer_blocks.2.attn.norm_add_k.weight, transformer_blocks.4.attn.norm_add_q.weight.
Please make sure to pass low_cpu_mem_usage=False and device_map=None if you want to randomly initialize those weights or else make sure your checkpoint file is correct.
Hi, I have the same problem
How do you deal with it?
I had this problem. I had the sd3 models downloaded and ran python app.py
I checked the source code and edited line 37 from
model_path = os.path.join(current_directory, "pyramid_flow_model") # Directory to store the model
to
model_path = os.path.join(current_directory, "pyramid-flow-miniflux") # Directory to store the model
The tried python app.py and tried an example. It started generating without errors so far. it's slow. I only have a RTX 3060 with 12gb of VRAM.
It had an error and ended the process. But it finds now the models.