Tensor size mismatch error during inference

[hardikdava]

Great work!, I am trying to run ACE++ for subject mode. I have used diffuser based pipeline version. I am getting following error. I have tried inputs as of same size but still getting same error. Any lead would be super helpful.

Thank you in advance.

0%|          | 0/28 [00:00<?, ?it/s]
0%|          | 0/28 [00:00<?, ?it/s]
Traceback (most recent call last):
File "/root/.pyenv/versions/3.11.10/lib/python3.11/site-packages/cog/server/worker.py", line 352, in _predict
result = predict(**payload)
^^^^^^^^^^^^^^^^^^
File "/root/.pyenv/versions/3.11.10/lib/python3.11/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/src/predictor.py", line 124, in predict
image = self.pipe(
^^^^^^^^^^
File "/root/.pyenv/versions/3.11.10/lib/python3.11/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/src/ace_plus_diffusers.py", line 105, in __call__
image = self.pipe(
^^^^^^^^^^
File "/root/.pyenv/versions/3.11.10/lib/python3.11/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/root/.pyenv/versions/3.11.10/lib/python3.11/site-packages/diffusers/pipelines/flux/pipeline_flux_fill.py", line 916, in __call__
noise_pred = self.transformer(
^^^^^^^^^^^^^^^^^
File "/root/.pyenv/versions/3.11.10/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1739, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/.pyenv/versions/3.11.10/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1750, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/.pyenv/versions/3.11.10/lib/python3.11/site-packages/diffusers/models/transformers/transformer_flux.py", line 463, in forward
hidden_states = self.x_embedder(hidden_states)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/.pyenv/versions/3.11.10/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1739, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/.pyenv/versions/3.11.10/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1750, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/.pyenv/versions/3.11.10/lib/python3.11/site-packages/torch/nn/modules/linear.py", line 125, in forward
return F.linear(input, self.weight, self.bias)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: mat1 and mat2 shapes cannot be multiplied (5504x384 and 64x3072)

[yaosheng216]

https://github.com/ali-vilab/ACE_plus/blob/main/inference/ace_plus_diffusers.py
@hardikdava You can refer of this code to used diffuser.

[hardikdava]

@yaosheng216 thanks for the reply. I have used the same code and am getting the above mentioned error.

[yaosheng216]

@yaosheng216 thanks for the reply. I have used the same code and am getting the above mentioned error.
@hardikdava Can you provide more error information.

[hardikdava]

@yaosheng216 I have used this script as starting point and used these images as an input. I have used default parameters only. I have not changed code in anyway. Actually, I am trying to deploy it on replicate. I have attached full stacktrace in the description itself.

[yaosheng216]

@hardikdava Used this script, you can invoke of screenshot's function, parameters of reference_image, edit_image, edit_mask, prompt, task and seed values set refer of this link: https://huggingface.co/spaces/scepter-studio/ACE-Plus. Another, this diffuser script is used FFT model.

[hardikdava]

@yaosheng216 Is it possible that transformers module used in diffuser workflow uses original transformer module? Doesn't it need to be replaced with modified transformer?

[yaosheng216]

@hardikdava Recommend you can try to upgrade diffusers and transformers to latest version.

[hardikdava]

@yaosheng216 I have used latest version of both package.

[yaosheng216]

@hardikdava Both of package version is no problem, can you provide more runtime information, this will helps me find the problem.

[yaosheng216]

@hardikdava comfyui_subject_lora16.safetensors is a LoRA model, this diffusers script need to used FFT model, that has same function with subject LoRA model, model url: https://huggingface.co/ali-vilab/ACE_Plus/blob/main/ace_plus_fft.safetensors

[hardikdava]

@yaosheng216 so which lora should I use for subject reconstruction problem but with diffuser pipeline?

[songyiren725]

Use the Flux-Fill-dev rather than Flux-dev

[songyiren725]

subject lora, the base model should be Flux-fill-dev

…

________________________________ From: Hardik Dava ***@***.***> Sent: Friday, March 28, 2025 8:14 PM To: ali-vilab/ACE_plus ***@***.***> Cc: SONG YIREN ***@***.***>; Comment ***@***.***> Subject: Re: [ali-vilab/ACE_plus] Tensor size mismatch error during inference (Issue #87) - External Email - @yaosheng216<https://ddec1-0-en-ctp.trendmicro.com:443/wis/clicktime/v1/query?url=https%3a%2f%2fgithub.com%2fyaosheng216&umid=6571d444-f059-404d-ad9a-044476b5f9fb&rct=1743164061&auth=8d3ccd473d52f326e51c0f75cb32c9541898e5d5-39ee88d17f786b558fcc045add8dc15ef41e33dd> so which lora should I use for subject reconstruction problem but with diffuser pipeline? — Reply to this email directly, view it on GitHub<https://ddec1-0-en-ctp.trendmicro.com:443/wis/clicktime/v1/query?url=https%3a%2f%2fgithub.com%2fali%2dvilab%2fACE%5fplus%2fissues%2f87%23issuecomment%2d2761190232&umid=6571d444-f059-404d-ad9a-044476b5f9fb&rct=1743164061&auth=8d3ccd473d52f326e51c0f75cb32c9541898e5d5-edc19775d542870d362e695ab6ec76d768f26fc6>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/BFCSILCTLA5LRZJY6KKV2LD2WUVI3AVCNFSM6AAAAABZ24VEVOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDONRRGE4TAMRTGI>. You are receiving this because you commented.Message ID: ***@***.***> [hardikdava]hardikdava left a comment (ali-vilab/ACE_plus#87)<https://ddec1-0-en-ctp.trendmicro.com:443/wis/clicktime/v1/query?url=https%3a%2f%2fgithub.com%2fali%2dvilab%2fACE%5fplus%2fissues%2f87%23issuecomment%2d2761190232&umid=6571d444-f059-404d-ad9a-044476b5f9fb&rct=1743164061&auth=8d3ccd473d52f326e51c0f75cb32c9541898e5d5-edc19775d542870d362e695ab6ec76d768f26fc6> @yaosheng216<https://ddec1-0-en-ctp.trendmicro.com:443/wis/clicktime/v1/query?url=https%3a%2f%2fgithub.com%2fyaosheng216&umid=6571d444-f059-404d-ad9a-044476b5f9fb&rct=1743164061&auth=8d3ccd473d52f326e51c0f75cb32c9541898e5d5-39ee88d17f786b558fcc045add8dc15ef41e33dd> so which lora should I use for subject reconstruction problem but with diffuser pipeline? — Reply to this email directly, view it on GitHub<https://ddec1-0-en-ctp.trendmicro.com:443/wis/clicktime/v1/query?url=https%3a%2f%2fgithub.com%2fali%2dvilab%2fACE%5fplus%2fissues%2f87%23issuecomment%2d2761190232&umid=6571d444-f059-404d-ad9a-044476b5f9fb&rct=1743164061&auth=8d3ccd473d52f326e51c0f75cb32c9541898e5d5-edc19775d542870d362e695ab6ec76d768f26fc6>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/BFCSILCTLA5LRZJY6KKV2LD2WUVI3AVCNFSM6AAAAABZ24VEVOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDONRRGE4TAMRTGI>. You are receiving this because you commented.Message ID: ***@***.***>

________________________________ Important: This email is confidential and may be privileged. If you are not the intended recipient, please delete it and notify us immediately; you should not copy or use it for any purpose, nor disclose its contents to any other person. Thank you.

[hardikdava]

@songyiren725 Not working. Still getting same error.

[yaosheng216]

https://openart.ai/workflows/whale_waterlogged_60/ace-more-convenient-replacement-of-everything/gjAsh5rGjfC6OEB2AUZv
@hardikdava Recommand you can try used workflow in the ComfyUI.