VAE Decoding Error when using MiniMax H3

[ReloadProcz]

Custom Node Testing

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

Your question

Whether using tiled vae or not, the error always show after the sampling process:

Traceback (most recent call last):
  File "/home/adam/comfyui/execution.py", line 545, 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 "/home/adam/comfyui/execution.py", line 344, 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 "/home/adam/comfyui/execution.py", line 318, in _async_map_node_over_list
    await process_inputs(input_dict, i)
  File "/home/adam/comfyui/execution.py", line 306, in process_inputs
    result = f(**inputs)
             ^^^^^^^^^^^
  File "/home/adam/comfyui/comfy_api/internal/__init__.py", line 149, in wrapped_func
    return method(locked_class, **inputs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/comfy_api/latest/_io.py", line 1935, in EXECUTE_NORMALIZED
    to_return = cls.execute(*args, **kwargs)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/comfy_extras/nodes_audio.py", line 156, in execute
    return IO.NodeOutput(vae_decode_audio(vae, samples, tile_size, overlap))
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/comfy_extras/nodes_audio.py", line 104, in vae_decode_audio
    audio = vae.decode_tiled(latent, tile_x=tile, tile_y=tile, overlap=overlap).movedim(-1, 1)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/comfy/sd.py", line 1239, in decode_tiled
    output = self.decode_tiled_(samples, **args)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/comfy/sd.py", line 1076, in decode_tiled_
    (comfy.utils.tiled_scale(samples, decode_fn, tile_x // 2, tile_y * 2, overlap, upscale_amount = self.upscale_ratio, output_device=self.output_device, pbar = pbar) +
     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/comfy/utils.py", line 1226, in tiled_scale
    return tiled_scale_multidim(samples, function, (tile_y, tile_x), overlap=overlap, upscale_amount=upscale_amount, out_channels=out_channels, output_device=output_device, pbar=pbar)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/venv/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/comfy/utils.py", line 1192, in tiled_scale_multidim
    ps = function(s_in).to(output_device)
         ^^^^^^^^^^^^^^
  File "/home/adam/comfyui/comfy/sd.py", line 1074, in <lambda>
    decode_fn = lambda a: self.first_stage_model.decode(a.to(self.vae_dtype).to(self.device)).to(dtype=self.vae_output_dtype())
                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/comfy/ldm/minimax/audio_vae.py", line 423, in decode
    x = self.decoder(x)  # [b * s, 1, L], already clamped to [-1, 1]
        ^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1786, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/comfy/ldm/minimax/audio_vae.py", line 362, in forward
    xs = self.resblocks[i * self.num_kernels + j](x)
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1786, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/comfy/ldm/minimax/audio_vae.py", line 299, in forward
    xt = a1(x)
         ^^^^^
  File "/home/adam/comfyui/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1786, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/comfy/ldm/minimax/audio_vae.py", line 146, in forward
    x = self.upsample(x)
        ^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1786, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/comfy/ldm/minimax/audio_vae.py", line 101, in forward
    x = F.conv_transpose1d(x, self.filter.expand(C, -1, -1).to(x.dtype), stride=self.stride, groups=C).mul_(self.ratio)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same

I removed out the audio decoding, only video decoding:

!!! Exception during processing !!! to() received an invalid combination of arguments - got (NestedTensor), but expected one of:
 * (torch.device device = None, torch.dtype dtype = None, bool non_blocking = False, bool copy = False, *, torch.memory_format memory_format = None)
 * (torch.dtype dtype, bool non_blocking = False, bool copy = False, *, torch.memory_format memory_format = None)
 * (Tensor tensor, bool non_blocking = False, bool copy = False, *, torch.memory_format memory_format = None)

[ERROR] Traceback (most recent call last):
  File "/home/adam/comfyui/execution.py", line 545, 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 "/home/adam/comfyui/execution.py", line 344, 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 "/home/adam/comfyui/execution.py", line 318, in _async_map_node_over_list
    await process_inputs(input_dict, i)
  File "/home/adam/comfyui/execution.py", line 306, in process_inputs
    result = f(**inputs)
             ^^^^^^^^^^^
  File "/home/adam/comfyui/nodes.py", line 368, in decode
    images = vae.decode_tiled(samples["samples"], tile_x=tile_size // compression, tile_y=tile_size // compression, overlap=overlap // compression, tile_t=temporal_size, overlap_t=temporal_overlap)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/comfy/sd.py", line 1234, in decode_tiled
    output = self._decode_tiled_owned(samples, **self._owned_tiled_args(tile_x, tile_y, overlap, tile_t, overlap_t))
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/comfy/sd.py", line 1097, in _decode_tiled_owned
    out = self.first_stage_model.decode_tiled(samples.to(self.vae_dtype).to(self.device), **kwargs)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/adam/comfyui/comfy/ldm/minimax/vae.py", line 678, in decode_tiled
    return self.decode(z)
           ^^^^^^^^^^^^^^
  File "/home/adam/comfyui/comfy/ldm/minimax/vae.py", line 682, in decode
    latents_mean = self.latents_mean.view(1, -1, 1, 1, 1).to(z)
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: to() received an invalid combination of arguments - got (NestedTensor), but expected one of:
 * (torch.device device = None, torch.dtype dtype = None, bool non_blocking = False, bool copy = False, *, torch.memory_format memory_format = None)
 * (torch.dtype dtype, bool non_blocking = False, bool copy = False, *, torch.memory_format memory_format = None)
 * (Tensor tensor, bool non_blocking = False, bool copy = False, *, torch.memory_format memory_format = None)

Startup parameters (excluding the custom node disabling): --novram --disable-cuda-malloc --disable-smart-memory --disable-pinned-memory --fast-disk --cache-classic

Logs

Other

The process should be fine when main.py has the startup parameter --lowvram or even --novram.
I tested it under --novram, lowvram takes too much VRAM for my Laptop, I don't have that patience lol

[coderabbitai]

🔗 Related PRs

#15210 - Support MiniMax H3 [open]
#15224 - feat: Support MiniMax-H3 (CORE-375) [open]

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!

[ReloadProcz]

Forgot to say the env I use:
WSL 2 Ubuntu 24.04 LTS (Be lazy to switch back to pure win11 lol)
venv with torch==2.9.1+cu128
(I know this may be not important at all)

[cushycrux]

I have a cryptic compile error after sampling:

100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [02:18<00:00,  6.90s/it]
[INFO] Requested to load MiniMaxH3AudioVAE
[INFO] Model MiniMaxH3AudioVAE prepared for dynamic VRAM loading. 576MB Staged. 0 patches attached. Force pre-loaded 401 weights: 539 KB.
[INFO] Model MiniMaxH3VideoVAE prepared for dynamic VRAM loading. 5309MB Staged. 0 patches attached. Force pre-loaded 128 weights: 352 KB.
C:/Users/Dave/AppData/Local/Temp/tmp_85veqko/cuda_utils.c:14: error: include file 'Python.h' not found
Failed to compile. cc_cmd: ['S:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\runtime\\tcc\\tcc.exe', 'C:\\Users\\Dave\\AppData\\Local\\Temp\\tmp_85veqko\\cuda_utils.c', '-O3', '-shared', '-Wno-psabi', '-o', 'C:\\Users\\Dave\\AppData\\Local\\Temp\\tmp_85veqko\\cuda_utils.cp313-win_amd64.pyd', '-fPIC', '-D_Py_USE_GCC_BUILTIN_ATOMICS', '-lcuda', '-lpython313', '-LS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\lib', '-LS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\lib\\x64', '-IS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IC:\\Users\\Dave\\AppData\\Local\\Temp\\tmp_85veqko', '-IS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Include']
[ERROR] !!! Exception during processing !!! Command '['S:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\runtime\\tcc\\tcc.exe', 'C:\\Users\\Dave\\AppData\\Local\\Temp\\tmp_85veqko\\cuda_utils.c', '-O3', '-shared', '-Wno-psabi', '-o', 'C:\\Users\\Dave\\AppData\\Local\\Temp\\tmp_85veqko\\cuda_utils.cp313-win_amd64.pyd', '-fPIC', '-D_Py_USE_GCC_BUILTIN_ATOMICS', '-lcuda', '-lpython313', '-LS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\lib', '-LS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\lib\\x64', '-IS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IC:\\Users\\Dave\\AppData\\Local\\Temp\\tmp_85veqko', '-IS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Include']' returned non-zero exit status 1.
[ERROR] Traceback (most recent call last):
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\execution.py", line 545, 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 "S:\AI\ComfyUI_windows_portable_26\ComfyUI\execution.py", line 344, 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 "S:\AI\ComfyUI_windows_portable_26\ComfyUI\execution.py", line 318, in _async_map_node_over_list
    await process_inputs(input_dict, i)
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\execution.py", line 306, in process_inputs
    result = f(**inputs)
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\nodes.py", line 335, in decode
    images = vae.decode(latent)
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\sd.py", line 1188, in decode
    model_management.raise_non_oom(e)
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\model_management.py", line 395, in raise_non_oom
    raise e
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\sd.py", line 1181, in decode
    out = self.first_stage_model.decode(samples, **vae_options).to(device=self.output_device, dtype=self.vae_output_dtype(), copy=True)
          ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\ldm\minimax\vae.py", line 690, in decode
    dec = self.decode_temporal(z)
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\ldm\minimax\vae.py", line 627, in decode_temporal
    clip_dec = self._adaptive_decode(clip_z)
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\ldm\minimax\vae.py", line 397, in _adaptive_decode
    return self.tiled_decode(z)
           ~~~~~~~~~~~~~~~~~^^^
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\ldm\minimax\vae.py", line 499, in tiled_decode
    tile = self._decode_pixels(z[..., zi:zi + zl, zj:zj + zw])
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\ldm\minimax\vae.py", line 388, in _decode_pixels
    return self.decoder(self.post_quant_conv(z))
           ~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1778, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1789, in _call_impl
    return forward_call(*args, **kwargs)
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\ldm\minimax\vae.py", line 301, in forward
    h = block(h, rotary_pos_emb)
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1778, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1789, in _call_impl
    return forward_call(*args, **kwargs)
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\ldm\minimax\vae.py", line 256, in forward
    x = x.addcmul_(self.attn(comfy.rmsnorm.rms_norm(x, self.norm1.weight, self.norm1.eps), rotary_pos_emb), self.scale1)
                   ~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1778, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1789, in _call_impl
    return forward_call(*args, **kwargs)
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\ldm\minimax\vae.py", line 236, in forward
    query[..., :rot], key[..., :rot] = comfy.quant_ops.ck.apply_rope_split_half(
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
        query[..., :rot], key[..., :rot], rotary_pos_emb)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\comfy_kitchen\__init__.py", line 642, in apply_rope_split_half
    return torch.ops.comfy_kitchen.apply_rope_split_half(xq, xk, freqs_cis)
           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\_ops.py", line 1275, in __call__
    return self._op(*args, **kwargs)
           ~~~~~~~~^^^^^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\_library\custom_ops.py", line 375, in backend_impl
    result = self._backend_fns[device_type](*args, **kwargs)
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\_compile.py", line 54, in inner
    return disable_fn(*args, **kwargs)
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\_dynamo\eval_frame.py", line 1298, in _fn
    return fn(*args, **kwargs)
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\_library\custom_ops.py", line 410, in wrapped_fn
    return fn(*args, **kwargs)
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\comfy_kitchen\backends\eager\rope.py", line 457, in _op_apply_rope_split_half
    return impl(**kwargs)
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\comfy_kitchen\backends\triton\rope.py", line 268, in apply_rope_split_half
    return _apply_rope(xq, freqs_cis, xk, split_half=True)
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\comfy_kitchen\backends\triton\rope.py", line 200, in _apply_rope
    apply_rope_kernel[grid](
    ~~~~~~~~~~~~~~~~~~~~~~~^
        x1,
        ^^^
    ...<27 lines>...
        split_half=split_half,
        ^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\triton\runtime\jit.py", line 419, in <lambda>
    return lambda *args, **kwargs: self.run(grid=grid, warmup=False, *args, **kwargs)
                                   ~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\triton\runtime\jit.py", line 713, in run
    device = driver.active.get_current_device()
             ^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\triton\runtime\driver.py", line 28, in active
    self._active = self.default
                   ^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\triton\runtime\driver.py", line 22, in default
    self._default = _create_driver()
                    ~~~~~~~~~~~~~~^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\triton\runtime\driver.py", line 10, in _create_driver
    return active_drivers[0]()
           ~~~~~~~~~~~~~~~~~^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\triton\backends\nvidia\driver.py", line 755, in __init__
    self.utils = CudaUtils()  # TODO: make static
                 ~~~~~~~~~^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\triton\backends\nvidia\driver.py", line 71, in __init__
    mod = compile_module_from_src(
        src=Path(os.path.join(dirname, "driver.c")).read_text(),
    ...<3 lines>...
        libraries=libraries,
    )
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\triton\runtime\build.py", line 169, in compile_module_from_src
    so = _build(name, src_path, tmpdir, library_dirs or [], include_dirs or [], libraries or [], ccflags or [])
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\triton\runtime\build.py", line 128, in _build
    raise e
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\triton\runtime\build.py", line 125, in _build
    subprocess.check_call(cc_cmd)
    ~~~~~~~~~~~~~~~~~~~~~^^^^^^^^
  File "subprocess.py", line 419, in check_call
subprocess.CalledProcessError: Command '['S:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\runtime\\tcc\\tcc.exe', 'C:\\Users\\Dave\\AppData\\Local\\Temp\\tmp_85veqko\\cuda_utils.c', '-O3', '-shared', '-Wno-psabi', '-o', 'C:\\Users\\Dave\\AppData\\Local\\Temp\\tmp_85veqko\\cuda_utils.cp313-win_amd64.pyd', '-fPIC', '-D_Py_USE_GCC_BUILTIN_ATOMICS', '-lcuda', '-lpython313', '-LS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\lib', '-LS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\lib\\x64', '-IS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IC:\\Users\\Dave\\AppData\\Local\\Temp\\tmp_85veqko', '-IS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Include']' returned non-zero exit status 1.

Whelp.

[patientx]

same error here, amd , rocm, windows

[ReloadProcz]

I have a cryptic compile error after sampling:

100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [02:18<00:00,  6.90s/it]
[INFO] Requested to load MiniMaxH3AudioVAE
[INFO] Model MiniMaxH3AudioVAE prepared for dynamic VRAM loading. 576MB Staged. 0 patches attached. Force pre-loaded 401 weights: 539 KB.
[INFO] Model MiniMaxH3VideoVAE prepared for dynamic VRAM loading. 5309MB Staged. 0 patches attached. Force pre-loaded 128 weights: 352 KB.
C:/Users/Dave/AppData/Local/Temp/tmp_85veqko/cuda_utils.c:14: error: include file 'Python.h' not found
Failed to compile. cc_cmd: ['S:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\runtime\\tcc\\tcc.exe', 'C:\\Users\\Dave\\AppData\\Local\\Temp\\tmp_85veqko\\cuda_utils.c', '-O3', '-shared', '-Wno-psabi', '-o', 'C:\\Users\\Dave\\AppData\\Local\\Temp\\tmp_85veqko\\cuda_utils.cp313-win_amd64.pyd', '-fPIC', '-D_Py_USE_GCC_BUILTIN_ATOMICS', '-lcuda', '-lpython313', '-LS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\lib', '-LS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\lib\\x64', '-IS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IC:\\Users\\Dave\\AppData\\Local\\Temp\\tmp_85veqko', '-IS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Include']
[ERROR] !!! Exception during processing !!! Command '['S:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\runtime\\tcc\\tcc.exe', 'C:\\Users\\Dave\\AppData\\Local\\Temp\\tmp_85veqko\\cuda_utils.c', '-O3', '-shared', '-Wno-psabi', '-o', 'C:\\Users\\Dave\\AppData\\Local\\Temp\\tmp_85veqko\\cuda_utils.cp313-win_amd64.pyd', '-fPIC', '-D_Py_USE_GCC_BUILTIN_ATOMICS', '-lcuda', '-lpython313', '-LS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\lib', '-LS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\lib\\x64', '-IS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IC:\\Users\\Dave\\AppData\\Local\\Temp\\tmp_85veqko', '-IS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Include']' returned non-zero exit status 1.
[ERROR] Traceback (most recent call last):
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\execution.py", line 545, 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 "S:\AI\ComfyUI_windows_portable_26\ComfyUI\execution.py", line 344, 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 "S:\AI\ComfyUI_windows_portable_26\ComfyUI\execution.py", line 318, in _async_map_node_over_list
    await process_inputs(input_dict, i)
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\execution.py", line 306, in process_inputs
    result = f(**inputs)
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\nodes.py", line 335, in decode
    images = vae.decode(latent)
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\sd.py", line 1188, in decode
    model_management.raise_non_oom(e)
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\model_management.py", line 395, in raise_non_oom
    raise e
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\sd.py", line 1181, in decode
    out = self.first_stage_model.decode(samples, **vae_options).to(device=self.output_device, dtype=self.vae_output_dtype(), copy=True)
          ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\ldm\minimax\vae.py", line 690, in decode
    dec = self.decode_temporal(z)
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\ldm\minimax\vae.py", line 627, in decode_temporal
    clip_dec = self._adaptive_decode(clip_z)
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\ldm\minimax\vae.py", line 397, in _adaptive_decode
    return self.tiled_decode(z)
           ~~~~~~~~~~~~~~~~~^^^
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\ldm\minimax\vae.py", line 499, in tiled_decode
    tile = self._decode_pixels(z[..., zi:zi + zl, zj:zj + zw])
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\ldm\minimax\vae.py", line 388, in _decode_pixels
    return self.decoder(self.post_quant_conv(z))
           ~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1778, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1789, in _call_impl
    return forward_call(*args, **kwargs)
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\ldm\minimax\vae.py", line 301, in forward
    h = block(h, rotary_pos_emb)
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1778, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1789, in _call_impl
    return forward_call(*args, **kwargs)
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\ldm\minimax\vae.py", line 256, in forward
    x = x.addcmul_(self.attn(comfy.rmsnorm.rms_norm(x, self.norm1.weight, self.norm1.eps), rotary_pos_emb), self.scale1)
                   ~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1778, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1789, in _call_impl
    return forward_call(*args, **kwargs)
  File "S:\AI\ComfyUI_windows_portable_26\ComfyUI\comfy\ldm\minimax\vae.py", line 236, in forward
    query[..., :rot], key[..., :rot] = comfy.quant_ops.ck.apply_rope_split_half(
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
        query[..., :rot], key[..., :rot], rotary_pos_emb)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\comfy_kitchen\__init__.py", line 642, in apply_rope_split_half
    return torch.ops.comfy_kitchen.apply_rope_split_half(xq, xk, freqs_cis)
           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\_ops.py", line 1275, in __call__
    return self._op(*args, **kwargs)
           ~~~~~~~~^^^^^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\_library\custom_ops.py", line 375, in backend_impl
    result = self._backend_fns[device_type](*args, **kwargs)
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\_compile.py", line 54, in inner
    return disable_fn(*args, **kwargs)
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\_dynamo\eval_frame.py", line 1298, in _fn
    return fn(*args, **kwargs)
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\torch\_library\custom_ops.py", line 410, in wrapped_fn
    return fn(*args, **kwargs)
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\comfy_kitchen\backends\eager\rope.py", line 457, in _op_apply_rope_split_half
    return impl(**kwargs)
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\comfy_kitchen\backends\triton\rope.py", line 268, in apply_rope_split_half
    return _apply_rope(xq, freqs_cis, xk, split_half=True)
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\comfy_kitchen\backends\triton\rope.py", line 200, in _apply_rope
    apply_rope_kernel[grid](
    ~~~~~~~~~~~~~~~~~~~~~~~^
        x1,
        ^^^
    ...<27 lines>...
        split_half=split_half,
        ^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\triton\runtime\jit.py", line 419, in <lambda>
    return lambda *args, **kwargs: self.run(grid=grid, warmup=False, *args, **kwargs)
                                   ~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\triton\runtime\jit.py", line 713, in run
    device = driver.active.get_current_device()
             ^^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\triton\runtime\driver.py", line 28, in active
    self._active = self.default
                   ^^^^^^^^^^^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\triton\runtime\driver.py", line 22, in default
    self._default = _create_driver()
                    ~~~~~~~~~~~~~~^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\triton\runtime\driver.py", line 10, in _create_driver
    return active_drivers[0]()
           ~~~~~~~~~~~~~~~~~^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\triton\backends\nvidia\driver.py", line 755, in __init__
    self.utils = CudaUtils()  # TODO: make static
                 ~~~~~~~~~^^
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\triton\backends\nvidia\driver.py", line 71, in __init__
    mod = compile_module_from_src(
        src=Path(os.path.join(dirname, "driver.c")).read_text(),
    ...<3 lines>...
        libraries=libraries,
    )
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\triton\runtime\build.py", line 169, in compile_module_from_src
    so = _build(name, src_path, tmpdir, library_dirs or [], include_dirs or [], libraries or [], ccflags or [])
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\triton\runtime\build.py", line 128, in _build
    raise e
  File "S:\AI\ComfyUI_windows_portable_26\python_embeded\Lib\site-packages\triton\runtime\build.py", line 125, in _build
    subprocess.check_call(cc_cmd)
    ~~~~~~~~~~~~~~~~~~~~~^^^^^^^^
  File "subprocess.py", line 419, in check_call
subprocess.CalledProcessError: Command '['S:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\runtime\\tcc\\tcc.exe', 'C:\\Users\\Dave\\AppData\\Local\\Temp\\tmp_85veqko\\cuda_utils.c', '-O3', '-shared', '-Wno-psabi', '-o', 'C:\\Users\\Dave\\AppData\\Local\\Temp\\tmp_85veqko\\cuda_utils.cp313-win_amd64.pyd', '-fPIC', '-D_Py_USE_GCC_BUILTIN_ATOMICS', '-lcuda', '-lpython313', '-LS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\lib', '-LS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\lib\\x64', '-IS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IC:\\Users\\Dave\\AppData\\Local\\Temp\\tmp_85veqko', '-IS:\\AI\\ComfyUI_windows_portable_26\\python_embeded\\Include']' returned non-zero exit status 1.

Whelp.

U Use triton-windows?

[ReloadProcz]

But for my problem, I updated torch to:
torch==2.11.0+cu128
So that in windows env I finally solved my problem... (Now testing WSL 2)

[cushycrux]

Just fixed my problem by doing this:
https://github.com/woct0rdho/triton-windows#8-special-notes-for-comfyui-with-embeded-python |o/
Now the video only has a wrong aspect ratio. -.-

[ReloadProcz]

Well on my WSL 2 still could not solve the problem only in video decoding, May be I rly have to use windows ones to do so... lol

[hum-ma]

I get the same TypeError about NestedTensor as OP but only when attempting tiled VAE decoding of video.

Linux, CUDA, torch 2.9.1, startup options --disable-smart-memory --cache-classic

Edit: ah, according to #15274 there is no point trying to use the tiled VAE node as the video VAE has built-in tiled decoding.