The size of tensor a (16) must match the size of tensor b (36) at non-singleton dimension 1

[putdanil]

Custom Node Testing

• I have tried disabling custom nodes and the issue persists (see how to disable custom nodes if you need help)

Expected Behavior

Sometimes I get this error with WAN 2.2 I2V 14B, use it with SageAttention and EasyCache. Feels like a problem with EasyCache.

Actual Behavior

Error: The size of tensor a (16) must match the size of tensor b (36) at non-singleton dimension 1

Steps to Reproduce

wan-native (14) (2).json

Debug Logs

File "/comfyui/comfy/samplers.py", line 963, in predict_noise
    return sampling_function(self.inner_model, x, timestep, self.conds.get("negative", None), self.conds.get("positive", None), self.cfg, model_options=model_options, seed=seed)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/comfyui/comfy/samplers.py", line 381, in sampling_function
    out = calc_cond_batch(model, conds, x, timestep, model_options)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/comfyui/comfy/samplers.py", line 206, in calc_cond_batch
    return _calc_cond_batch_outer(model, conds, x_in, timestep, model_options)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/comfyui/comfy/samplers.py", line 214, in _calc_cond_batch_outer
    return executor.execute(model, conds, x_in, timestep, model_options)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/comfyui/comfy/patcher_extension.py", line 113, in execute
    return self.wrappers[self.idx](self, *args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/comfyui/comfy_extras/nodes_easycache.py", line 142, in easycache_calc_cond_batch_wrapper
    return executor(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/comfyui/comfy/patcher_extension.py", line 105, in call
    return new_executor.execute(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/comfyui/comfy/patcher_extension.py", line 112, in execute
    return self.original(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/comfyui/comfy/samplers.py", line 326, in _calc_cond_batch
    output = model.apply_model(input_x, timestep_, **c).chunk(batch_chunks)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/comfyui/comfy/model_base.py", line 161, in apply_model
    return comfy.patcher_extension.WrapperExecutor.new_class_executor(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/comfyui/comfy/patcher_extension.py", line 112, in execute
    return self.original(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/comfyui/comfy/model_base.py", line 207, in _apply_model
    return self.model_sampling.calculate_denoised(sigma, model_output.float(), x)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/comfyui/comfy/model_sampling.py", line 68, in calculate_denoised
    return model_input - model_output * sigma
           ~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~
RuntimeError: The size of tensor a (16) must match the size of tensor b (36) at non-singleton dimension 1

2025-10-24T12:35:50.396319 - Prompt executed in 56.61 seconds

Other

No response

[mrfrosty009]

Most likely its the problem with the resolution, it should be divisible by 16. Try it, I might be mistaken.

[thrnz]

I've had the same error, though I also spotted EasyCache throwing the following warning recently as well. I'm not sure if the two are related:

/ComfyUI/comfy_extras/nodes_easycache.py:286: UserWarning: Using a non-tuple sequence for multidimensional indexing is deprecated and will be changed in pytorch 2.9; use x[tuple(seq)] instead of x[seq]. In pytorch 2.9 this will be interpreted as tensor index, x[torch.tensor(seq)], which will result either in an error or a different result (Triggered internally at /pytorch/torch/csrc/autograd/python_variable_indexing.cpp:345.)

I've tried the suggested tweak to line 286 in nodes_easycache.py and am currently waiting to see if the error still reproduces.

[thrnz]

I've tried the suggested tweak to line 286 in nodes_easycache.py and am currently waiting to see if the error still reproduces.

Well that was quick. Unfortunately that change doesn't help.

[putdanil]

I've tried the suggested tweak to line 286 in nodes_easycache.py and am currently waiting to see if the error still reproduces.

Well that was quick. Unfortunately that change doesn't help.

It's not a solution, but looks like setting resolution divisible by 36 helps. I don't get why sometimes it causes an error, and sometimes it doesn't.

[thrnz]

Enabling EasyCache's verbose logging seems to suggest that it triggers when a step is skipped. Setting start percent to 0.00 and reuse_threshold to 3.00 to force an early skip seems to be enough to reliably trigger it early on.

[2025-10-25 21:40:42.714] EasyCache [verbose] - skipping step; cumulative_change_rate: 0.029388427734375, reuse_threshold: 0.04
[2025-10-25 21:40:42.729] EasyCache [verbose] - output_change_rates 1: [0.025390625]
[2025-10-25 21:40:42.730] EasyCache [verbose] - approx_output_change_rates 0: []
[2025-10-25 21:40:42.731] EasyCache - skipped 1/10 steps (1.11x speedup).
[2025-10-25 21:40:42.755] !!! Exception during processing !!! The size of tensor a (16) must match the size of tensor b (36) at non-singleton dimension 1
...
[2025-10-25 21:48:37.396] EasyCache [verbose] - skipping step; cumulative_change_rate: 0.0699462890625, reuse_threshold: 3.0
[2025-10-25 21:48:37.397] 
[2025-10-25 21:48:37.398] EasyCache [verbose] - output_change_rates 1: [0.049224853515625]
[2025-10-25 21:48:37.398] EasyCache [verbose] - approx_output_change_rates 0: []
[2025-10-25 21:48:37.398] EasyCache - skipped 1/10 steps (1.11x speedup).
[2025-10-25 21:48:37.421] !!! Exception during processing !!! The size of tensor a (16) must match the size of tensor b (36) at non-singleton dimension 1

[thrnz]

Reverting #10376 seems to fix it.

[thrnz]

Looks like it might've gotten addressed in #10788

[Kosinkadink]

Yep, was an issue on models were input channels did not match output channels. Should now be fixed