{{ message }}
Repository navigation
Qwen 2.1 Image Edit workflow crashes on a Blackwell GPU when xformers is installed #17198
Description
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
No crash.
Actual Behavior
[ERROR] !!! Exception during processing !!! No operator found for memory_efficient_attention_forward with inputs:
Expand
[ERROR] !!! Exception during processing !!! No operator found for `memory_efficient_attention_forward` with inputs:
query : shape=(1, 8, 32, 128) (torch.bfloat16)
key : shape=(1, 8, 32, 128) (torch.bfloat16)
value : shape=(1, 8, 32, 128) (torch.bfloat16)
attn_bias :
p : 0.0
`fa3F@0.0.0` is not supported because:
requires device with capability <= (9, 0) but your GPU has capability (12, 0) (too new)
attn_bias type is
operator wasn't built - see `python -m xformers.info` for more info
`fa2F@2.8.3.post1` is not supported because:
attn_bias type is
`cutlassF-pt` is not supported because:
requires device with capability <= (9, 0) but your GPU has capability (12, 0) (too new)
[ERROR] Traceback (most recent call last):
File "ComfyUI/execution.py", line 548, 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 "ComfyUI/execution.py", line 353, 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 "ComfyUI/execution.py", line 327, in _async_map_node_over_list
await process_inputs(input_dict, i)
File "ComfyUI/execution.py", line 315, in process_inputs
result = f(**inputs)
File "ComfyUI/nodes.py", line 1633, in sample
return common_ksampler(model, seed, steps, cfg, sampler_name, scheduler, positive, negative, latent_image, denoise=denoise)
File "ComfyUI/nodes.py", line 1597, in common_ksampler
samples = comfy.sample.sample(model, noise, steps, cfg, sampler_name, scheduler, positive, negative, latent_image,
denoise=denoise, disable_noise=disable_noise, start_step=start_step, last_step=last_step,
force_full_denoise=force_full_denoise, noise_mask=noise_mask, callback=callback, disable_pbar=disable_pbar, seed=seed)
File "ComfyUI/comfy/sample.py", line 82, in sample
samples = sampler.sample(noise, positive, negative, cfg=cfg, latent_image=latent_image, start_step=start_step, last_step=last_step, force_full_denoise=force_full_denoise, denoise_mask=noise_mask, sigmas=sigmas, callback=callback, disable_pbar=disable_pbar, seed=seed)
File "ComfyUI/comfy/samplers.py", line 1463, in sample
return sample(self.model, noise, positive, negative, cfg, self.device, sampler, sigmas, self.model_options, latent_image=latent_image, denoise_mask=denoise_mask, callback=callback, disable_pbar=disable_pbar, seed=seed)
File "ComfyUI/comfy/samplers.py", line 1353, in sample
return cfg_guider.sample(noise, latent_image, sampler, sigmas, denoise_mask, callback, disable_pbar, seed)
~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "ComfyUI/comfy/samplers.py", line 1335, in sample
output = executor.execute(noise, latent_image, sampler, sigmas, denoise_mask, callback, disable_pbar, seed, latent_shapes=latent_shapes)
File "ComfyUI/comfy/patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "ComfyUI/comfy/samplers.py", line 1262, in outer_sample
output = self.inner_sample(noise, latent_image, device, sampler, sigmas, denoise_mask, callback, disable_pbar, seed, latent_shapes=latent_shapes)
File "ComfyUI/comfy/samplers.py", line 1237, in inner_sample
samples = executor.execute(self, sigmas, extra_args, callback, noise, latent_image, denoise_mask, disable_pbar)
File "ComfyUI/comfy/patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "ComfyUI/comfy/samplers.py", line 1005, in sample
samples = self.sampler_function(model_k, noise, sigmas, extra_args=extra_args, callback=k_callback, disable=disable_pbar, **self.extra_options)
File "ComfyUI/venv/lib/python3.14/site-packages/torch/utils/_contextlib.py", line 124, in decorate_context
return func(*args, **kwargs)
File "ComfyUI/comfy/k_diffusion/sampling.py", line 205, in sample_euler
denoised = model(x, sigma_hat * s_in, **extra_args)
File "ComfyUI/comfy/samplers.py", line 640, in __call__
out = self.inner_model(x, sigma, model_options=model_options, seed=seed)
File "ComfyUI/comfy/samplers.py", line 1208, in __call__
return self.outer_predict_noise(*args, **kwargs)
~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "ComfyUI/comfy/samplers.py", line 1215, in outer_predict_noise
).execute(x, timestep, model_options, seed)
~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "ComfyUI/comfy/patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "ComfyUI/comfy/samplers.py", line 1218, 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 620, in sampling_function
out = calc_cond_batch(model, conds, x, timestep, model_options)
File "ComfyUI/comfy/samplers.py", line 211, in calc_cond_batch
return _calc_cond_batch_outer(model, conds, x_in, timestep, model_options)
File "ComfyUI/comfy/samplers.py", line 219, 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.original(*args, **kwargs)
~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "ComfyUI/comfy/samplers.py", line 335, in _calc_cond_batch
output = model.apply_model(input_x, timestep_, **c).chunk(batch_chunks)
~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^
File "ComfyUI/comfy/model_base.py", line 212, in apply_model
return comfy.patcher_extension.WrapperExecutor.new_class_executor(
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
...<2 lines>...
comfy.patcher_extension.get_all_wrappers(comfy.patcher_extension.WrappersMP.APPLY_MODEL, transformer_options)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
).execute(x, t, c_concat, c_crossattn, control, transformer_options, **kwargs)
~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "ComfyUI/comfy/patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "ComfyUI/comfy/model_base.py", line 256, in _apply_model
model_output = self.diffusion_model(xc, t, context=context, control=control, transformer_options=transformer_options, **extra_conds)
File "ComfyUI/venv/lib/python3.14/site-packages/torch/nn/modules/module.py", line 1778, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "ComfyUI/venv/lib/python3.14/site-packages/torch/nn/modules/module.py", line 1789, in _call_impl
return forward_call(*args, **kwargs)
File "ComfyUI/comfy/ldm/qwen_image21/model.py", line 342, in forward
return comfy.patcher_extension.WrapperExecutor.new_class_executor(
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
...<2 lines>...
comfy.patcher_extension.get_all_wrappers(comfy.patcher_extension.WrappersMP.DIFFUSION_MODEL, transformer_options)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
).execute(x, timestep, context, ref_latents, image_slots, transformer_options, **kwargs)
~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "ComfyUI/comfy/patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "ComfyUI/comfy/ldm/qwen_image21/model.py", line 393, in _forward
prefix_states = block(prefix_states, mod, prefix_pe, prefix_attn, prefix_states.shape[1], transformer_options)
File "ComfyUI/venv/lib/python3.14/site-packages/torch/nn/modules/module.py", line 1778, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "ComfyUI/venv/lib/python3.14/site-packages/torch/nn/modules/module.py", line 1789, in _call_impl
return forward_call(*args, **kwargs)
File "ComfyUI/comfy/ldm/qwen_image21/model.py", line 146, in forward
x = _gated_residual(x, self.attn(_modulated_norm(self.img_norm1, x, scale1, prefix_len, zero), pe, attn_fn, prefix_len, transformer_options), gate1, prefix_len)
~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "ComfyUI/venv/lib/python3.14/site-packages/torch/nn/modules/module.py", line 1778, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "ComfyUI/venv/lib/python3.14/site-packages/torch/nn/modules/module.py", line 1789, in _call_impl
return forward_call(*args, **kwargs)
File "ComfyUI/comfy/ldm/qwen_image21/model.py", line 108, in forward
return self.to_out[0](attn_fn(q, k, v, self.heads, preferred_attention=self.comfy_attention))
~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "ComfyUI/comfy/ldm/qwen_image21/model.py", line 211, in attn
outs.append(optimized_attention(query, key, value, heads, mask=mask, transformer_options=transformer_options, preferred_attention=preferred_attention))
~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "ComfyUI/comfy/ldm/modules/attention.py", line 248, in wrapper
return func(*args, **kwargs)
File "ComfyUI/comfy/ldm/modules/attention.py", line 576, in attention_xformers
out = xformers.ops.memory_efficient_attention(q, k, v, attn_bias=mask, scale=kwargs.get("scale", None))
File "ComfyUI/venv/lib/python3.14/site-packages/xformers/ops/fmha/__init__.py", line 311, in memory_efficient_attention
return _memory_efficient_attention(
Inputs(
...<8 lines>...
op=op,
)
File "ComfyUI/venv/lib/python3.14/site-packages/xformers/ops/fmha/__init__.py", line 472, in _memory_efficient_attention
return _memory_efficient_attention_forward(
inp, op=op[0] if op is not None else None
)
File "ComfyUI/venv/lib/python3.14/site-packages/xformers/ops/fmha/__init__.py", line 491, in _memory_efficient_attention_forward
op = _dispatch_fw(inp, False)
File "ComfyUI/venv/lib/python3.14/site-packages/xformers/ops/fmha/dispatch.py", line 142, in _dispatch_fw
return _run_priority_list(
"memory_efficient_attention_forward",
_dispatch_fw_priority_list(inp, needs_gradient),
inp,
)
File "ComfyUI/venv/lib/python3.14/site-packages/xformers/ops/fmha/dispatch.py", line 83, in _run_priority_list
raise NotImplementedError(msg)
NotImplementedError: No operator found for `memory_efficient_attention_forward` with inputs:
query : shape=(1, 8, 32, 128) (torch.bfloat16)
key : shape=(1, 8, 32, 128) (torch.bfloat16)
value : shape=(1, 8, 32, 128) (torch.bfloat16)
attn_bias :
p : 0.0
`fa3F@0.0.0` is not supported because:
requires device with capability <= (9, 0) but your GPU has capability (12, 0) (too new)
attn_bias type is
operator wasn't built - see `python -m xformers.info` for more info
`fa2F@2.8.3.post1` is not supported because:
attn_bias type is
`cutlassF-pt` is not supported because:
requires device with capability <= (9, 0) but your GPU has capability (12, 0) (too new)
Steps to Reproduce
- Install
xformers. - Run this workflow on a Blackwell GPU: https://paste.sh/4Vaade_F#IlF6HagS4kSazCa9zOxN5ASV
Debug Logs
$ python main.py --disable-all-custom-nodes
[INFO] setup plugin alembic.autogenerate.schemas
[INFO] setup plugin alembic.autogenerate.tables
[INFO] setup plugin alembic.autogenerate.types
[INFO] setup plugin alembic.autogenerate.constraints
[INFO] setup plugin alembic.autogenerate.defaults
[INFO] setup plugin alembic.autogenerate.comments
[INFO] Found comfy_kitchen backend eager: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['adaln', 'apply_rope', 'apply_rope1', 'apply_rope1_', 'apply_rope_', 'apply_rope_split_half', 'apply_rope_split_half1', 'apply_rope_split_half1_', 'apply_rope_split_half_', 'convrot_w4a4_linear', 'dequantize_convrot_w4a4_weight', 'dequantize_int8_convrot_weight', 'dequantize_int8_convrot_weight_dtype', 'dequantize_int8_embedding', 'dequantize_int8_simple', 'dequantize_int8_simple_dtype', 'dequantize_mxfp8', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'dequantize_w4a8_int8_weight', 'fp16_conv3d', 'fp16_conv3d_out', 'fp16_linear', 'gemv_awq_w4a16', 'group_norm_silu_pad3d', 'group_norm_silu_pad3d_out', 'int8_linear', 'na3d', 'prepare_int4_weight_for_int8_linear', 'quantize_and_rotate_rowwise', 'quantize_convrot_w4a4_weight', 'quantize_int8_convrot_weight', 'quantize_int8_rowwise', 'quantize_int8_tensorwise', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'quantize_svdquant_w4a4', 'quantize_w4a8_int8_weight', 'rms_adaln', 'rms_rope', 'rms_rope1', 'rms_rope1_', 'rms_rope_', 'rms_rope_split_half', 'rms_rope_split_half1', 'rms_rope_split_half1_', 'rms_rope_split_half_', 'rotate_int8_convrot_weight', 'scaled_mm_mxfp8', 'scaled_mm_nvfp4', 'scaled_mm_svdquant_w4a4', 'sol_attn', 'stochastic_rounding_fp8', 'w4a8_int8_linear', 'w4a8_packed_linear']}
[INFO] Found comfy_kitchen backend hip: {'available': False, 'disabled': False, 'unavailable_reason': 'PyTorch ROCm/HIP runtime not available', 'capabilities': []}
[INFO] Found comfy_kitchen backend ascend: {'available': False, 'disabled': False, 'unavailable_reason': "torch-npu is not installed: No module named 'torch_npu'", 'capabilities': []}
[INFO] Found comfy_kitchen backend cuda: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['adaln', 'apply_rope', 'apply_rope1', 'apply_rope1_', 'apply_rope_', 'apply_rope_split_half', 'apply_rope_split_half1', 'apply_rope_split_half1_', 'apply_rope_split_half_', 'closest_point_on_mesh', 'connected_components', 'convrot_w4a4_linear', 'delaunay3d', 'dequantize_convrot_w4a4_weight', 'dequantize_int8_convrot_weight', 'dequantize_int8_convrot_weight_dtype', 'dequantize_int8_simple', 'dequantize_int8_simple_dtype', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'dequantize_w4a8_int8_weight', 'edge_collapse_checks', 'fp16_conv3d', 'fp16_conv3d_out', 'fp16_linear', 'gemv_awq_w4a16', 'group_norm_silu_pad3d', 'group_norm_silu_pad3d_out', 'int8_linear', 'mesh_bvh', 'min_cut', 'na3d', 'prepare_int4_weight_for_int8_linear', 'quantize_and_rotate_rowwise', 'quantize_convrot_w4a4_weight', 'quantize_int8_convrot_weight', 'quantize_int8_rowwise', 'quantize_int8_tensorwise', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'quantize_svdquant_w4a4', 'quantize_w4a8_int8_weight', 'rms_adaln', 'rms_rope', 'rms_rope1', 'rms_rope1_', 'rms_rope_', 'rms_rope_split_half', 'rms_rope_split_half1', 'rms_rope_split_half1_', 'rms_rope_split_half_', 'rotate_int8_convrot_weight', 'scaled_mm_nvfp4', 'scaled_mm_svdquant_w4a4', 'sol_attn', 'stochastic_rounding_fp8', 'w4a8_int8_linear', 'w4a8_packed_linear']}
[INFO] Found comfy_kitchen backend triton: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['adaln', 'apply_rope', 'apply_rope1', 'apply_rope1_', 'apply_rope_', 'apply_rope_split_half', 'apply_rope_split_half1', 'apply_rope_split_half1_', 'apply_rope_split_half_', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'int8_linear', 'na3d', 'quantize_and_rotate_rowwise', 'quantize_int8_rowwise', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'rms_adaln', 'rms_rope', 'rms_rope1', 'rms_rope1_', 'rms_rope_', 'rms_rope_split_half', 'rms_rope_split_half1', 'rms_rope_split_half1_', 'rms_rope_split_half_', 'w4a8_int8_linear']}
[INFO] Checkpoint files will always be loaded safely.
[INFO] Total VRAM 24079 MB, total RAM 80228 MB
[INFO] pytorch version: 2.12.1+cu130
[INFO] xformers version: 0.0.35
[INFO] Set vram state to: NORMAL_VRAM
[INFO] Device: cuda:0 NVIDIA RTX PRO 4000 Blackwell : cudaMallocAsync
[INFO] Device: cuda:1 NVIDIA GeForce RTX 5060 Ti : cudaMallocAsync
[INFO] Using async weight offloading with 2 streams
[INFO] Enabled pinned memory 63844
[INFO] Using xformers attention
[INFO] aimdo: /project/src/control.c:283:INFO:comfy-aimdo inited for GPU: NVIDIA RTX PRO 4000 Blackwell (VRAM: 24079 MB)
[INFO] aimdo: /project/src/control.c:283:INFO:comfy-aimdo inited for GPU: NVIDIA GeForce RTX 5060 Ti (VRAM: 15932 MB)
[INFO] DynamicVRAM support detected and enabled
[INFO] Python version: 3.14.7 (main, Sep 26 2026, 18:07:20) [GCC 14.3.1 20260604]
[INFO] ComfyUI version: 0.42.0
[INFO] comfy-aimdo version: 0.5.6
[INFO] comfy-kitchen version: 0.2.39
[INFO] comfyui-frontend-package version: 1.55.18
[INFO] comfyui-workflow-templates version: 0.11.89
[INFO] comfyui-embedded-docs version: 0.5.14
[INFO] comfy-kitchen version: 0.2.39
[INFO] comfy-aimdo version: 0.5.6
[INFO] [Prompt Server] web root: /workspace/ComfyUI/venv/lib/python3.14/site-packages/comfyui_frontend_package/static
[INFO] Asset seeder disabled
[INFO] No OpenGL_accelerate module loaded: No module named 'OpenGL_accelerate'
[INFO] Skipping loading of custom nodes
[INFO] Using RAM pressure cache.
[INFO] Starting server
[INFO] To see the GUI go to: http://127.0.0.1:8188
[INFO] got prompt
[INFO] Model storage policy: fast_disk=True paths=['/workspace/ComfyUI/models/vae/qwen_image_2.1_vae_bf16.safetensors']
[INFO] Using xformers attention in VAE
[INFO] Using xformers attention in VAE
[INFO] VAE load device: cuda:0, offload device: cpu, dtype: torch.bfloat16
[INFO] Found quantization metadata version 1
[INFO] Using MixedPrecisionOps for text encoder
[INFO] Model storage policy: fast_disk=True paths=['/workspace/ComfyUI/models/text_encoders/qwen3vl_8b_int8_convrot.safetensors']
[INFO] CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
[INFO] Requested to load WanVAE
[INFO] Model WanVAE prepared for dynamic VRAM loading. 644MB Staged. 0 patches attached. Force pre-loaded 68 weights: 102 KB.
[INFO] Requested to load QwenImage21TEModel_
[INFO] Model QwenImage21TEModel_ prepared for dynamic VRAM loading. 8916MB Staged. 0 patches attached. Force pre-loaded 255 weights: 849 KB.
[INFO] Model QwenImage21TEModel_ prepared for dynamic VRAM loading. 8916MB Staged. 0 patches attached. Force pre-loaded 255 weights: 849 KB.
[INFO] Found quantization metadata version 1
[INFO] Detected mixed precision quantization
[INFO] Using mixed precision operations
[INFO] Native ops: nvfp4, mxfp8, convrot_w4a4, int8_tensorwise, w6a8_int8, float8_e5m2, asym_w4a8_int8, float8_e4m3fn
[INFO] model weight dtype torch.bfloat16, manual cast: torch.bfloat16
[INFO] model_type FLUX
[INFO] Model storage policy: fast_disk=True paths=['/workspace/ComfyUI/models/diffusion_models/qwen_image_2.1_int8_convrot.safetensors']
[INFO] Requested to load QwenImage21
[INFO] Model QwenImage21 prepared for dynamic VRAM loading. 6920MB Staged. 0 patches attached. Force pre-loaded 65 weights: 63 KB.
0%| | 0/25 [00:00<?, ?it/s, Model Initializing ... ]
[INFO] Comfy model compiler graph breaks: 0, rogues: 0
[ERROR] !!! Exception during processing !!! No operator found for `memory_efficient_attention_forward` with inputs:
query : shape=(1, 8, 32, 128) (torch.bfloat16)
key : shape=(1, 8, 32, 128) (torch.bfloat16)
value : shape=(1, 8, 32, 128) (torch.bfloat16)
attn_bias : <class 'torch.Tensor'>
p : 0.0
`fa3F@0.0.0` is not supported because:
requires device with capability <= (9, 0) but your GPU has capability (12, 0) (too new)
attn_bias type is <class 'torch.Tensor'>
operator wasn't built - see `python -m xformers.info` for more info
`fa2F@2.8.3.post1` is not supported because:
attn_bias type is <class 'torch.Tensor'>
`cutlassF-pt` is not supported because:
requires device with capability <= (9, 0) but your GPU has capability (12, 0) (too new)
[ERROR] Traceback (most recent call last):
File "/workspace/ComfyUI/execution.py", line 548, 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 353, 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 327, in _async_map_node_over_list
await process_inputs(input_dict, i)
File "/workspace/ComfyUI/execution.py", line 315, in process_inputs
result = f(**inputs)
File "/workspace/ComfyUI/nodes.py", line 1633, in sample
return common_ksampler(model, seed, steps, cfg, sampler_name, scheduler, positive, negative, latent_image, denoise=denoise)
File "/workspace/ComfyUI/nodes.py", line 1597, in common_ksampler
samples = comfy.sample.sample(model, noise, steps, cfg, sampler_name, scheduler, positive, negative, latent_image,
denoise=denoise, disable_noise=disable_noise, start_step=start_step, last_step=last_step,
force_full_denoise=force_full_denoise, noise_mask=noise_mask, callback=callback, disable_pbar=disable_pbar, seed=seed)
File "/workspace/ComfyUI/comfy/sample.py", line 82, in sample
samples = sampler.sample(noise, positive, negative, cfg=cfg, latent_image=latent_image, start_step=start_step, last_step=last_step, force_full_denoise=force_full_denoise, denoise_mask=noise_mask, sigmas=sigmas, callback=callback, disable_pbar=disable_pbar, seed=seed)
File "/workspace/ComfyUI/comfy/samplers.py", line 1463, in sample
return sample(self.model, noise, positive, negative, cfg, self.device, sampler, sigmas, self.model_options, latent_image=latent_image, denoise_mask=denoise_mask, callback=callback, disable_pbar=disable_pbar, seed=seed)
File "/workspace/ComfyUI/comfy/samplers.py", line 1353, in sample
return cfg_guider.sample(noise, latent_image, sampler, sigmas, denoise_mask, callback, disable_pbar, seed)
~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/comfy/samplers.py", line 1335, in sample
output = executor.execute(noise, latent_image, sampler, sigmas, denoise_mask, callback, disable_pbar, seed, latent_shapes=latent_shapes)
File "/workspace/ComfyUI/comfy/patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/comfy/samplers.py", line 1262, in outer_sample
output = self.inner_sample(noise, latent_image, device, sampler, sigmas, denoise_mask, callback, disable_pbar, seed, latent_shapes=latent_shapes)
File "/workspace/ComfyUI/comfy/samplers.py", line 1237, in inner_sample
samples = executor.execute(self, sigmas, extra_args, callback, noise, latent_image, denoise_mask, disable_pbar)
File "/workspace/ComfyUI/comfy/patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/comfy/samplers.py", line 1005, in sample
samples = self.sampler_function(model_k, noise, sigmas, extra_args=extra_args, callback=k_callback, disable=disable_pbar, **self.extra_options)
File "/workspace/ComfyUI/venv/lib/python3.14/site-packages/torch/utils/_contextlib.py", line 124, in decorate_context
return func(*args, **kwargs)
File "/workspace/ComfyUI/comfy/k_diffusion/sampling.py", line 205, in sample_euler
denoised = model(x, sigma_hat * s_in, **extra_args)
File "/workspace/ComfyUI/comfy/samplers.py", line 640, in __call__
out = self.inner_model(x, sigma, model_options=model_options, seed=seed)
File "/workspace/ComfyUI/comfy/samplers.py", line 1208, in __call__
return self.outer_predict_noise(*args, **kwargs)
~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/comfy/samplers.py", line 1215, in outer_predict_noise
).execute(x, timestep, model_options, seed)
~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/comfy/patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/comfy/samplers.py", line 1218, 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 "/workspace/ComfyUI/comfy/samplers.py", line 620, in sampling_function
out = calc_cond_batch(model, conds, x, timestep, model_options)
File "/workspace/ComfyUI/comfy/samplers.py", line 211, in calc_cond_batch
return _calc_cond_batch_outer(model, conds, x_in, timestep, model_options)
File "/workspace/ComfyUI/comfy/samplers.py", line 219, in _calc_cond_batch_outer
return executor.execute(model, conds, x_in, timestep, model_options)
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/comfy/patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/comfy/samplers.py", line 335, in _calc_cond_batch
output = model.apply_model(input_x, timestep_, **c).chunk(batch_chunks)
~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/comfy/model_base.py", line 212, in apply_model
return comfy.patcher_extension.WrapperExecutor.new_class_executor(
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
...<2 lines>...
comfy.patcher_extension.get_all_wrappers(comfy.patcher_extension.WrappersMP.APPLY_MODEL, transformer_options)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
).execute(x, t, c_concat, c_crossattn, control, transformer_options, **kwargs)
~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/comfy/patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/comfy/model_base.py", line 256, in _apply_model
model_output = self.diffusion_model(xc, t, context=context, control=control, transformer_options=transformer_options, **extra_conds)
File "/workspace/ComfyUI/venv/lib/python3.14/site-packages/torch/nn/modules/module.py", line 1778, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/venv/lib/python3.14/site-packages/torch/nn/modules/module.py", line 1789, in _call_impl
return forward_call(*args, **kwargs)
File "/workspace/ComfyUI/comfy/ldm/qwen_image21/model.py", line 342, in forward
return comfy.patcher_extension.WrapperExecutor.new_class_executor(
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
...<2 lines>...
comfy.patcher_extension.get_all_wrappers(comfy.patcher_extension.WrappersMP.DIFFUSION_MODEL, transformer_options)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
).execute(x, timestep, context, ref_latents, image_slots, transformer_options, **kwargs)
~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/comfy/patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/comfy/ldm/qwen_image21/model.py", line 393, in _forward
prefix_states = block(prefix_states, mod, prefix_pe, prefix_attn, prefix_states.shape[1], transformer_options)
File "/workspace/ComfyUI/venv/lib/python3.14/site-packages/torch/nn/modules/module.py", line 1778, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/venv/lib/python3.14/site-packages/torch/nn/modules/module.py", line 1789, in _call_impl
return forward_call(*args, **kwargs)
File "/workspace/ComfyUI/comfy/ldm/qwen_image21/model.py", line 146, in forward
x = _gated_residual(x, self.attn(_modulated_norm(self.img_norm1, x, scale1, prefix_len, zero), pe, attn_fn, prefix_len, transformer_options), gate1, prefix_len)
~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/venv/lib/python3.14/site-packages/torch/nn/modules/module.py", line 1778, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/venv/lib/python3.14/site-packages/torch/nn/modules/module.py", line 1789, in _call_impl
return forward_call(*args, **kwargs)
File "/workspace/ComfyUI/comfy/ldm/qwen_image21/model.py", line 108, in forward
return self.to_out[0](attn_fn(q, k, v, self.heads, preferred_attention=self.comfy_attention))
~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/comfy/ldm/qwen_image21/model.py", line 211, in attn
outs.append(optimized_attention(query, key, value, heads, mask=mask, transformer_options=transformer_options, preferred_attention=preferred_attention))
~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/workspace/ComfyUI/comfy/ldm/modules/attention.py", line 248, in wrapper
return func(*args, **kwargs)
File "/workspace/ComfyUI/comfy/ldm/modules/attention.py", line 576, in attention_xformers
out = xformers.ops.memory_efficient_attention(q, k, v, attn_bias=mask, scale=kwargs.get("scale", None))
File "/workspace/ComfyUI/venv/lib/python3.14/site-packages/xformers/ops/fmha/__init__.py", line 311, in memory_efficient_attention
return _memory_efficient_attention(
Inputs(
...<8 lines>...
op=op,
)
File "/workspace/ComfyUI/venv/lib/python3.14/site-packages/xformers/ops/fmha/__init__.py", line 472, in _memory_efficient_attention
return _memory_efficient_attention_forward(
inp, op=op[0] if op is not None else None
)
File "/workspace/ComfyUI/venv/lib/python3.14/site-packages/xformers/ops/fmha/__init__.py", line 491, in _memory_efficient_attention_forward
op = _dispatch_fw(inp, False)
File "/workspace/ComfyUI/venv/lib/python3.14/site-packages/xformers/ops/fmha/dispatch.py", line 142, in _dispatch_fw
return _run_priority_list(
"memory_efficient_attention_forward",
_dispatch_fw_priority_list(inp, needs_gradient),
inp,
)
File "/workspace/ComfyUI/venv/lib/python3.14/site-packages/xformers/ops/fmha/dispatch.py", line 83, in _run_priority_list
raise NotImplementedError(msg)
NotImplementedError: No operator found for `memory_efficient_attention_forward` with inputs:
query : shape=(1, 8, 32, 128) (torch.bfloat16)
key : shape=(1, 8, 32, 128) (torch.bfloat16)
value : shape=(1, 8, 32, 128) (torch.bfloat16)
attn_bias : <class 'torch.Tensor'>
p : 0.0
`fa3F@0.0.0` is not supported because:
requires device with capability <= (9, 0) but your GPU has capability (12, 0) (too new)
attn_bias type is <class 'torch.Tensor'>
operator wasn't built - see `python -m xformers.info` for more info
`fa2F@2.8.3.post1` is not supported because:
attn_bias type is <class 'torch.Tensor'>
`cutlassF-pt` is not supported because:
requires device with capability <= (9, 0) but your GPU has capability (12, 0) (too new)Other
What an AI says:
Bug:
attention_xformers(comfy/ldm/modules/attention.py:526) passes Qwen‑Image 2.1's block‑causal mask straight into xformers as a denseattn_bias(line 576). xformers has no kernel for tensorattn_biason consumer Blackwell (capability 12.0) —cutlassF-ptcaps at (9,0), fa2/fa3 reject tensor bias — so dispatch throws. Selection is unconditional:optimized_attention_masked = optimized_attention(line 953), and Qwen‑Image 2.1 feeds per‑segment masks viablock_causal_attention(comfy/ldm/qwen_image21/model.py:211).Every other mask‑incapable backend already guards this (
attention_flashraises→SDPA,attention_sageline 717,attention_comfy_kitchen_solline 700); xformers is the only one missing the fallback.Secondary bug in the same function: the mask is bool, and
mask_outis built withdtype=q.dtype(line 569), so True→1.0/False→0.0 becomes an additive bias — nothing is masked (and allowed logits get +1). On sm80/90, where xformers "works", Qwen‑Image 2.1 edit silently ignores its causal mask.Fix: in
attention_xformers, addif mask is not None: return attention_pytorch(q, k, v, heads, mask, skip_reshape=skip_reshape, skip_output_reshape=skip_output_reshape, **kwargs)before the xformers call.
Activity
[-]Qwen 2.1 Image Edit workflow crashes on a Blackwell GPU[/-][+]Qwen 2.1 Image Edit workflow crashes on a Blackwell GPU when xformers is installed[/+]