Skip to content

Specified t5xxl_fp16.safetensors and it still wanted to download and use t5xxl_enconly.safetensors #407

Closed
Closed
Specified t5xxl_fp16.safetensors and it still wanted to download and use t5xxl_enconly.safetensors#407
Labels
BugSomething isn't working

Description

@Jonseed
Jonseed
opened on Nov 11, 2024

Expected Behavior

I expected it to use the t5xxl_fp16.safetensors I already have in text encoders folder, and that I specified in the UI (I turned on the T5xxl model switch, and chose the t5xxl_fp16.safetensors from the dropdown menu).

Actual Behavior

When I clicked generate, the logs say they didn't find the t5xxl_enconly.safetensors and it is currently downloading it.

Steps to Reproduce

Explained above. Why didn't it use the t5xxl_fp16 I specified in the UI?

Debug Logs

!!! Exception during processing !!! No such file or directory: "D:/repos/stable-diffusion-webui/models\text_encoder\t5xxl_enconly.safetensors"
2024-11-10 19:19:37.710 [Warning] [ComfyUI-2/STDERR] Traceback (most recent call last):
2024-11-10 19:19:37.711 [Warning] [ComfyUI-2/STDERR] File "D:\repos\ComfyUI\execution.py", line 323, in execute
2024-11-10 19:19:37.711 [Warning] [ComfyUI-2/STDERR] 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)
2024-11-10 19:19:37.712 [Warning] [ComfyUI-2/STDERR] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2024-11-10 19:19:37.713 [Warning] [ComfyUI-2/STDERR] File "D:\repos\ComfyUI\execution.py", line 198, in get_output_data
2024-11-10 19:19:37.714 [Warning] [ComfyUI-2/STDERR] 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)
2024-11-10 19:19:37.714 [Warning] [ComfyUI-2/STDERR] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2024-11-10 19:19:37.715 [Warning] [ComfyUI-2/STDERR] File "D:\repos\ComfyUI\execution.py", line 169, in _map_node_over_list
2024-11-10 19:19:37.716 [Warning] [ComfyUI-2/STDERR] process_inputs(input_dict, i)
2024-11-10 19:19:37.717 [Warning] [ComfyUI-2/STDERR] File "D:\repos\ComfyUI\execution.py", line 158, in process_inputs
2024-11-10 19:19:37.717 [Warning] [ComfyUI-2/STDERR] results.append(getattr(obj, func)(**inputs))
2024-11-10 19:19:37.718 [Warning] [ComfyUI-2/STDERR] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2024-11-10 19:19:37.719 [Warning] [ComfyUI-2/STDERR] File "D:\repos\ComfyUI\nodes.py", line 937, in load_clip
2024-11-10 19:19:37.720 [Warning] [ComfyUI-2/STDERR] clip = comfy.sd.load_clip(ckpt_paths=[clip_path1, clip_path2], embedding_directory=folder_paths.get_folder_paths("embeddings"), clip_type=clip_type)
2024-11-10 19:19:37.721 [Warning] [ComfyUI-2/STDERR] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2024-11-10 19:19:37.721 [Warning] [ComfyUI-2/STDERR] File "D:\repos\ComfyUI\comfy\sd.py", line 437, in load_clip
2024-11-10 19:19:37.722 [Warning] [ComfyUI-2/STDERR] clip_data.append(comfy.utils.load_torch_file(p, safe_load=True))
2024-11-10 19:19:37.723 [Warning] [ComfyUI-2/STDERR] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2024-11-10 19:19:37.723 [Warning] [ComfyUI-2/STDERR] File "D:\repos\ComfyUI\comfy\utils.py", line 34, in load_torch_file
2024-11-10 19:19:37.724 [Warning] [ComfyUI-2/STDERR] sd = safetensors.torch.load_file(ckpt, device=device.type)
2024-11-10 19:19:37.725 [Warning] [ComfyUI-2/STDERR] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2024-11-10 19:19:37.725 [Warning] [ComfyUI-2/STDERR] File "D:\repos\conda-envs\comfy\Lib\site-packages\safetensors\torch.py", line 313, in load_file
2024-11-10 19:19:37.726 [Warning] [ComfyUI-2/STDERR] with safe_open(filename, framework="pt", device=device) as f:
2024-11-10 19:19:37.727 [Warning] [ComfyUI-2/STDERR] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2024-11-10 19:19:37.727 [Warning] [ComfyUI-2/STDERR] FileNotFoundError: No such file or directory: "D:/repos/stable-diffusion-webui/models\text_encoder\t5xxl_enconly.safetensors"
2024-11-10 19:19:37.728 [Warning] [ComfyUI-2/STDERR]
2024-11-10 19:19:37.729 [Debug] [ComfyUI-2/STDERR] Prompt executed in 5.13 seconds
2024-11-10 19:19:38.003 [Error] [BackendHandler] backend #2 failed to load model with error: ComfyUI execution error: No such file or directory: "D:/repos/stable-diffusion-webui/models\text_encoder\t5xxl_enconly.safetensors"
2024-11-10 19:19:38.005 [Warning] [BackendHandler] backend #2 failed to load model flux/flux1-dev-Q8_0.gguf
2024-11-10 19:19:38.006 [Debug] Will deny backends: 2
2024-11-10 19:19:38.432 [Warning] [BackendHandler] All backends failed to load the model 'D:/repos/stable-diffusion-webui/models/diffusion_models/flux\flux1-dev-Q8_0.gguf'! Cannot generate anything.
2024-11-10 19:19:38.435 [Error] [BackendHandler] Backend request #1 failed: All available backends failed to load the model 'D:/repos/stable-diffusion-webui/models/diffusion_models/flux\flux1-dev-Q8_0.gguf'.
2024-11-10 19:19:38.437 [Error] [BackendHandler] Backend request #1 failed: All available backends failed to load the model 'D:/repos/stable-diffusion-webui/models/diffusion_models/flux\flux1-dev-Q8_0.gguf'.
2024-11-10 19:19:38.439 [Debug] [BackendHandler] Backend request #1 finished.
2024-11-10 19:19:38.440 [Debug] Refused to generate image for local: All available backends failed to load the model 'D:/repos/stable-diffusion-webui/models/diffusion_models/flux\flux1-dev-Q8_0.gguf'.
2024-11-10 19:19:58.924 [Info] t5xxl_enconly.safetensors download at 5.0%...

Other

If I import the workflow into the Comfy Workflow tab, it shows my t5xxl_fp16.safetensors in the dual clip loader. So I'm not sure why it is defaulting to this other t5xxl_enconly.safetensors file that I don't already have.

Activity

  1. Jonseed
    added
    BugSomething isn't working
    on Nov 11, 2024

  2. Jonseed commented on Nov 11, 2024

    @Jonseed
    Jonseed
    on Nov 11, 2024
    Author

    If I queue the job from the Comfy Workflow tab, it works fine, using the t5xxl_fp16 file specified.

  3. Jonseed commented on Nov 26, 2024

    @Jonseed
    Jonseed
    on Nov 26, 2024
    Author

    I'm still getting this error, where it wants to download t5xxl_enconly even though I've specified t5xxl_fp16 in the Generate tab of the UI (and enabled it). I've tried specifying other t5 quants, and they don't work either.

  4. mcmonkey4eva commented on Nov 26, 2024

    @mcmonkey4eva
    mcmonkey4eva
    on Nov 26, 2024
    Member

    Yes it's a valid bug, it's just low priority. The correct file you specified is used, it just also requires enconly be valid and usable for preload. It's low priority because this only affects people who are intentionally overfilling their ram with fp16, ie people who won't care if swarm is accidentally using a bit of extra resources on the enconly file too. If you care about resource consumption you'll naturally prefer the file that isn't twice the size for no reason.

  5. Jonseed commented on Nov 27, 2024

    @Jonseed
    Jonseed
    on Nov 27, 2024
    Author

    The problem is that it won't even generate an image without the enconly file present. It has to download the entire 4.7gb enconly file before generating anything when I have a valid fp16 t5 already specified (or any other t5).

    I'm not sure what you mean about overfilling ram with fp16 "for no reason," since fp16 produces more accurate images than fp8:
    city96/ComfyUI-GGUF#49 (comment)

  6. DocShotgun commented on Dec 7, 2024

    @DocShotgun
    DocShotgun
    on Dec 7, 2024

    I got around this by creating a symlink called t5xxl_enconly.safetensors that points to my pre-existing t5xxl. Not ideal but it works without taking up extra disk space I suppose.

  7. mcmonkey4eva
    added a commit that references this issue on Jan 7, 2025

    copy model params in the loader call

    9672fbb

  8. mcmonkey4eva commented on Jan 7, 2025

    @mcmonkey4eva
    mcmonkey4eva
    on Jan 7, 2025
    Member

    Swarm now will try to use your selected textenc models when loading for normal gens. If you trigger a contextless load (eg hit the "Load Model Now") button, it will still want to grab default models. You should still ideally have a model named t5xxl_enconly to serve as global default.

  9. mcmonkey4eva
    closed this as completedon Jan 7, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    BugSomething isn't working

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions