WanVideoSampler Error when using sageattn in infinitetalk workflow

Open

#12,053 opened on Jan 24, 2026 1 comment 0 reactions 0 assignees View on GitHub

Potential Bug

Dominant language: Python
Stars: 133k
Forks: 15.7k
Avg merge: 1d 7h
Merged PRs (30d): 158

Description

### Custom Node Testing

- [ ] I have tried disabling custom nodes and the issue persists (see [how to disable custom nodes](https://docs.comfy.org/troubleshooting/custom-node-issues#step-1%3A-test-with-all-custom-nodes-disabled) if you need help)

### Expected Behavior

WanVideoSampler should generate video frames normally when using SageAttention

### Actual Behavior

WanVideoSampler crashes immediately when processing begins with SageAttention enabled. The error occurs in SageAttention's CUDA kernel (transpose_pad_permute_cuda) when attempting to access tensor data.

### Steps to Reproduce

Update ComfyUI to include commit 16b9aab or later
Enable SageAttention (version 2.2.0)
Run a workflow using WanVideoSampler
Observe crash with "Cannot access data pointer" error

### Debug Logs

```powershell
Transformer weights loaded:
Device: cuda:0 | Memory: 4,277,416.86 MB
Device: cpu | Memory: 10,978,487.25 MB
Using 1261 LoRA weight patches for WanVideo model
------- Scheduler info -------
Total timesteps: tensor([999, 970, 916, 785], device='cuda:0')
Using timesteps: tensor([999, 970, 916, 785], device='cuda:0')
Using sigmas: tensor([1.0000, 0.9706, 0.9167, 0.7857, 0.0000])
------------------------------
Multitalk audio features shapes (per speaker): [(37, 12, 768)]
Rope function: comfy
Multitalk mode: infinitetalk
Sampling 37 frames in 1 windows, at 576x1024 with 4 steps
Sampling audio indices 0-81: 0%| | 0/4 [00:00, std::allocator >) + 0x9d (0x7303946ddfdd in /opt/venv/lib/python3.12/site-packages/torch/lib/libc10.so)
frame #1: c10::detail::torchCheckFail(char const*, char const*, unsigned int, char const*) + 0x69 (0x730394673637 in /opt/venv/lib/python3.12/site-packages/torch/lib/libc10.so)
frame #2: c10::TensorImpl::throw_data_ptr_access_error() const + 0x34 (0x7303946b7764 in /opt/venv/lib/python3.12/site-packages/torch/lib/libc10.so)
frame #3: void* c10::TensorImpl::data_impl(c10::TensorImpl::mutable_data()::{lambda()#1} const&) const + 0x98 (0x72ff88d1c3c8 in /opt/venv/lib/python3.12/site-packages/sageattention/_fused.cpython-312-x86_64-linux-gnu.so)
frame #4: quant_per_warp_int8_cuda(at::Tensor, at::Tensor, at::Tensor, int, int, int) + 0x2b42 (0x72ff88d15492 in /opt/venv/lib/python3.12/site-packages/sageattention/_fused.cpython-312-x86_64-linux-gnu.so)
frame #5: + 0x39d90 (0x72ff88d2cd90 in /opt/venv/lib/python3.12/site-packages/sageattention/_fused.cpython-312-x86_64-linux-gnu.so)
frame #6: + 0x372f7 (0x72ff88d2a2f7 in /opt/venv/lib/python3.12/site-packages/sageattention/_fused.cpython-312-x86_64-linux-gnu.so)
frame #7: python() [0x581a6f]

frame #10: + 0xa49f36 (0x7302caa04f36 in /opt/venv/lib/python3.12/site-packages/torch/lib/libtorch_python.so)
frame #11: + 0xdc0d85 (0x7302cad7bd85 in /opt/venv/lib/python3.12/site-packages/torch/lib/libtorch_python.so)
frame #12: + 0x66a755c (0x7302bb1f855c in /opt/venv/lib/python3.12/site-packages/torch/lib/libtorch_cpu.so)
frame #13: + 0xb13864 (0x7302caace864 in /opt/venv/lib/python3.12/site-packages/torch/lib/libtorch_python.so)
frame #14: + 0xb13d68 (0x7302caaced68 in /opt/venv/lib/python3.12/site-packages/torch/lib/libtorch_python.so)
frame #15: torch::jit::_get_operation_for_overload_or_packet(std::vector, std::allocator > > const&, c10::Symbol, pybind11::args const&, pybind11::kwargs const&, bool, std::optional) + 0x38 (0x7302caacf078 in /opt/venv/lib/python3.12/site-packages/torch/lib/libtorch_python.so)
frame #16: + 0x9f7a5d (0x7302ca9b2a5d in /opt/venv/lib/python3.12/site-packages/torch/lib/libtorch_python.so)
frame #17: + 0x40e440 (0x7302ca3c9440 in /opt/venv/lib/python3.12/site-packages/torch/lib/libtorch_python.so)
frame #18: python() [0x581a6f]
frame #22: python() [0x5a30c8]
frame #25: python() [0x54cf04]
frame #28: python() [0x54cf04]
frame #32: python() [0x5a30c8]
frame #35: python() [0x54cf04]
frame #38: python() [0x54cf04]
frame #42: python() [0x5a30c8]
frame #45: python() [0x54cf04]
frame #48: python() [0x54cf04]
frame #52: python() [0x5a30c8]
frame #55: python() [0x54cf04]
frame #58: python() [0x66b079]
frame #59: + 0x9d44 (0x7303985ccd44 in /usr/lib/python3.12/lib-dynload/_asyncio.cpython-312-x86_64-linux-gnu.so)
frame #60: + 0x9b5b (0x7303985ccb5b in /usr/lib/python3.12/lib-dynload/_asyncio.cpython-312-x86_64-linux-gnu.so)
frame #62: python() [0x6a3e1c]

Error during sampling: Cannot access data pointer of Tensor that doesn't have storage
Exception raised from throw_data_ptr_access_error at /pytorch/c10/core/TensorImpl.cpp:307 (most recent call first):
frame #0: c10::Error::Error(c10::SourceLocation, std::__cxx11::basic_string, std::allocator >) + 0x9d (0x7303946ddfdd in /opt/venv/lib/python3.12/site-packages/torch/lib/libc10.so)
frame #1: c10::detail::torchCheckFail(char const*, char const*, unsigned int, char const*) + 0x69 (0x730394673637 in /opt/venv/lib/python3.12/site-packages/torch/lib/libc10.so)
frame #2: c10::TensorImpl::throw_data_ptr_access_error() const + 0x34 (0x7303946b7764 in /opt/venv/lib/python3.12/site-packages/torch/lib/libc10.so)
frame #3: void* c10::TensorImpl::data_impl(c10::TensorImpl::mutable_data()::{lambda()#1} const&) const + 0x98 (0x72ff88d1c3c8 in /opt/venv/lib/python3.12/site-packages/sageattention/_fused.cpython-312-x86_64-linux-gnu.so)
frame #4: quant_per_warp_int8_cuda(at::Tensor, at::Tensor, at::Tensor, int, int, int) + 0x2b42 (0x72ff88d15492 in /opt/venv/lib/python3.12/site-packages/sageattention/_fused.cpython-312-x86_64-linux-gnu.so)
frame #5: + 0x39d90 (0x72ff88d2cd90 in /opt/venv/lib/python3.12/site-packages/sageattention/_fused.cpython-312-x86_64-linux-gnu.so)
frame #6: + 0x372f7 (0x72ff88d2a2f7 in /opt/venv/lib/python3.12/site-packages/sageattention/_fused.cpython-312-x86_64-linux-gnu.so)
frame #7: python() [0x581a6f]

!!! Exception during processing !!! Cannot access data pointer of Tensor that doesn't have storage
Exception raised from throw_data_ptr_access_error at /pytorch/c10/core/TensorImpl.cpp:307 (most recent call first):
frame #0: c10::Error::Error(c10::SourceLocation, std::__cxx11::basic_string, std::allocator >) + 0x9d (0x7303946ddfdd in /opt/venv/lib/python3.12/site-packages/torch/lib/libc10.so)
frame #1: c10::detail::torchCheckFail(char const*, char const*, unsigned int, char const*) + 0x69 (0x730394673637 in /opt/venv/lib/python3.12/site-packages/torch/lib/libc10.so)
frame #2: c10::TensorImpl::throw_data_ptr_access_error() const + 0x34 (0x7303946b7764 in /opt/venv/lib/python3.12/site-packages/torch/lib/libc10.so)
frame #3: void* c10::TensorImpl::data_impl(c10::TensorImpl::mutable_data()::{lambda()#1} const&) const + 0x98 (0x72ff88d1c3c8 in /opt/venv/lib/python3.12/site-packages/sageattention/_fused.cpython-312-x86_64-linux-gnu.so)
frame #4: quant_per_warp_int8_cuda(at::Tensor, at::Tensor, at::Tensor, int, int, int) + 0x2b42 (0x72ff88d15492 in /opt/venv/lib/python3.12/site-packages/sageattention/_fused.cpython-312-x86_64-linux-gnu.so)
frame #5: + 0x39d90 (0x72ff88d2cd90 in /opt/venv/lib/python3.12/site-packages/sageattention/_fused.cpython-312-x86_64-linux-gnu.so)
frame #6: + 0x372f7 (0x72ff88d2a2f7 in /opt/venv/lib/python3.12/site-packages/sageattention/_fused.cpython-312-x86_64-linux-gnu.so)
frame #7: python() [0x581a6f]

Traceback (most recent call last):
File "/workspace/ComfyUI/execution.py", line 518, in execute
output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/execution.py", line 329, in get_output_data
return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/execution.py", line 303, in _async_map_node_over_list
await process_inputs(input_dict, i)
File "/workspace/ComfyUI/execution.py", line 291, in process_inputs
result = f(**inputs)
^^^^^^^^^^^
File "/workspace/ComfyUI/custom_nodes/ComfyUI-WanVideoWrapper/nodes_sampler.py", line 2590, in process
raise e
File "/workspace/ComfyUI/custom_nodes/ComfyUI-WanVideoWrapper/nodes_sampler.py", line 2043, in process
return multitalk_loop(**locals())
^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/custom_nodes/ComfyUI-WanVideoWrapper/multitalk/multitalk_loop.py", line 339, in multitalk_loop
noise_pred, _, self.cache_state = predict_func(
^^^^^^^^^^^^^
File "/workspace/ComfyUI/custom_nodes/ComfyUI-WanVideoWrapper/nodes_sampler.py", line 1663, in predict_with_cfg
raise e
File "/workspace/ComfyUI/custom_nodes/ComfyUI-WanVideoWrapper/nodes_sampler.py", line 1510, in predict_with_cfg
noise_pred_cond, noise_pred_ovi, cache_state_cond = transformer(
^^^^^^^^^^^^
File "/opt/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1776, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1787, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/custom_nodes/ComfyUI-WanVideoWrapper/wanvideo/modules/model.py", line 3274, in forward
x, x_ip, lynx_ref_feature, x_ovi = block(x, x_ip=x_ip, lynx_ref_feature=lynx_ref_feature, x_ovi=x_ovi, x_onetoall_ref=x_onetoall_ref, onetoall_freqs=onetoall_freqs, attention_mode_override=attention_mode, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1776, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1787, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/custom_nodes/ComfyUI-WanVideoWrapper/wanvideo/modules/model.py", line 1346, in forward
x_audio = self.audio_cross_attn(self.norm_x(x.to(self.norm_x.weight.dtype)).to(input_dtype), encoder_hidden_states=multitalk_audio_embedding,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1776, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1787, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/custom_nodes/ComfyUI-WanVideoWrapper/multitalk/multitalk.py", line 320, in forward
return super().forward(x, encoder_hidden_states, shape)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/custom_nodes/ComfyUI-WanVideoWrapper/multitalk/multitalk.py", line 256, in forward
x = attention(q, encoder_k, encoder_v, attention_mode=self.attention_mode)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/custom_nodes/ComfyUI-WanVideoWrapper/wanvideo/modules/attention.py", line 109, in attention
return sageattn_func(q, k, v, tensor_layout="NHD").contiguous()
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/venv/lib/python3.12/site-packages/torch/_ops.py", line 1209, in __call__
return self._op(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/venv/lib/python3.12/site-packages/torch/_library/custom_ops.py", line 347, in backend_impl
result = self._backend_fns[device_type](*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/venv/lib/python3.12/site-packages/torch/_compile.py", line 54, in inner
return disable_fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/venv/lib/python3.12/site-packages/torch/_dynamo/eval_frame.py", line 1181, in _fn
return fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File "/opt/venv/lib/python3.12/site-packages/torch/_library/custom_ops.py", line 382, in wrapped_fn
return fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/custom_nodes/ComfyUI-WanVideoWrapper/wanvideo/modules/attention.py", line 24, in sageattn_func
return sageattn(q, k, v, attn_mask=attn_mask, dropout_p=dropout_p, is_causal=is_causal, tensor_layout=tensor_layout)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/venv/lib/python3.12/site-packages/sageattention/core.py", line 150, in sageattn
return sageattn_qk_int8_pv_fp8_cuda(q, k, v, tensor_layout=tensor_layout, is_causal=is_causal, qk_quant_gran="per_warp", sm_scale=sm_scale, return_lse=return_lse, pv_accum_dtype="fp32+fp16") # sm120 has accurate fp32 accumulator for fp8 mma and triton kernel is currently not usable on sm120.
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/venv/lib/python3.12/site-packages/torch/_dynamo/eval_frame.py", line 1181, in _fn
return fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File "/opt/venv/lib/python3.12/site-packages/sageattention/core.py", line 737, in sageattn_qk_int8_pv_fp8_cuda
q_int8, q_scale, k_int8, k_scale = per_warp_int8_cuda(q, k, km, tensor_layout=tensor_layout, BLKQ=128, WARPQ=32, BLKK=64)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/venv/lib/python3.12/site-packages/sageattention/quant.py", line 172, in per_warp_int8
_fused.quant_per_warp_int8_cuda(q, q_int8, q_scale, BLKQ, WARPQ, _tensor_layout)
RuntimeError: Cannot access data pointer of Tensor that doesn't have storage
Exception raised from throw_data_ptr_access_error at /pytorch/c10/core/TensorImpl.cpp:307 (most recent call first):
frame #0: c10::Error::Error(c10::SourceLocation, std::__cxx11::basic_string, std::allocator >) + 0x9d (0x7303946ddfdd in /opt/venv/lib/python3.12/site-packages/torch/lib/libc10.so)
frame #1: c10::detail::torchCheckFail(char const*, char const*, unsigned int, char const*) + 0x69 (0x730394673637 in /opt/venv/lib/python3.12/site-packages/torch/lib/libc10.so)
frame #2: c10::TensorImpl::throw_data_ptr_access_error() const + 0x34 (0x7303946b7764 in /opt/venv/lib/python3.12/site-packages/torch/lib/libc10.so)
frame #3: void* c10::TensorImpl::data_impl(c10::TensorImpl::mutable_data()::{lambda()#1} const&) const + 0x98 (0x72ff88d1c3c8 in /opt/venv/lib/python3.12/site-packages/sageattention/_fused.cpython-312-x86_64-linux-gnu.so)
frame #4: quant_per_warp_int8_cuda(at::Tensor, at::Tensor, at::Tensor, int, int, int) + 0x2b42 (0x72ff88d15492 in /opt/venv/lib/python3.12/site-packages/sageattention/_fused.cpython-312-x86_64-linux-gnu.so)
frame #5: + 0x39d90 (0x72ff88d2cd90 in /opt/venv/lib/python3.12/site-packages/sageattention/_fused.cpython-312-x86_64-linux-gnu.so)
frame #6: + 0x372f7 (0x72ff88d2a2f7 in /opt/venv/lib/python3.12/site-packages/sageattention/_fused.cpython-312-x86_64-linux-gnu.so)
frame #7: python() [0x581a6f]

Prompt executed in 14.47 seconds
Sampling audio indices 0-81: 0%| | 0/4 [00:01

Workaround
Use sdpa attention mode instead of sage in WanVideoSampler

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

WanVideoSampler Error when using sageattn in infinitetalk workflow

Description

Contributor guide

Assessment

Get new issues in your inbox