"norm_cuda" not implemented for 'Float8_e4m3fn'

[GalaxyTimeMachine]

I can't run your example workflow without an error. I'm using latest ComfyUI portable and get this every time:

!!! Exception during processing !!! "norm_cuda" not implemented for 'Float8_e4m3fn'
Traceback (most recent call last):
  File "D:\ComfyUI_windows_portable\ComfyUI\execution.py", line 327, in execute
    output_data, output_ui, has_subgraph = get_output_data(obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb)
                                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\ComfyUI_windows_portable\ComfyUI\execution.py", line 202, in get_output_data
    return_values = _map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\ComfyUI_windows_portable\ComfyUI\execution.py", line 174, in _map_node_over_list
    process_inputs(input_dict, i)
  File "D:\ComfyUI_windows_portable\ComfyUI\execution.py", line 163, in process_inputs
    results.append(getattr(obj, func)(**inputs))
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\ComfyUI_windows_portable\ComfyUI\custom_nodes\vector_sculptor_comfyui\nodes.py", line 164, in exec
    sculptor_tokens = vector_sculptor_tokens(clip, text, sculptor_method, token_normalization, sculptor_intensity)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\ComfyUI_windows_portable\ComfyUI\custom_nodes\vector_sculptor_comfyui\nodes.py", line 107, in vector_sculptor_tokens
    new_vector, n_found = refine_token_weight(token_id,all_weights, sculptor_method, actual_multiplier)
                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\ComfyUI_windows_portable\ComfyUI\custom_nodes\vector_sculptor_comfyui\nodes.py", line 33, in refine_token_weight
    pre_mag = torch.norm(initial_weight)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\functional.py", line 1805, in norm
    return torch.linalg.vector_norm(
           ^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: "norm_cuda" not implemented for 'Float8_e4m3fn'

[Extraltodeus]

Hello! Which model do you use?

[GalaxyTimeMachine]

Hi!, SDXL.

[Extraltodeus]

If I google it I find this which mentions a need to update torch. But what surprises me is that I have never seen what seems to be an 8bits quant for the clip g or l to use with SDXL. I may need more details if you don't feel like updating.

[GalaxyTimeMachine]

"...if you don't feel like updating" - I update the UI and nodes every day. What did you want me to try?

[GalaxyTimeMachine]

I have a working SDXL workflow. I replace the positive clip text encoder node with the node shown below, and then it breaks the workflow and I get the error.

[henrikvilhelmberglund]

A lot later, but it seems like this happens when using --fp8_e4m3fn-text-enc --fp8_e4m3fn-unet.