CLIPTextEncodeFlux very slow

[mno1337]

Your question

CLIPTextEncodeFlux its taking around 80-90 seconds, started a few days ago. Before that it didnt take so long. No vram issues or anything, latest version of comfy etc...

Logs

No response

Other

No response

[mno1337]

Ok it got fixed with updating pytorch to 2.5.0 but then sampler got twice as slow instead.. from 1.1s/it to 2.2s/it...

[Zdeto]

Same issue here, CLIPTextEncodeFlux taking painfully long times in the last few days. Didn't update pytorch to 2.5.0...

[ltdrdata]

What is your workflow and what is your hardware spec?

[Zdeto]

4ltData.json comfyui.log

All right, I've attached the workflow and the comfyui log. The first run of the workflow took 444s, the next run just ~250s, because the cliptext was already encoded. This workflow has a Facedetailer in it, but the prompt I run generated an image without discernable faces, so this step was basically skipped. But I used to run this workflow (including the Facedetailer steps) in under 300s!

My hardware specs... i7-10700K, 32Gb RAM, RTX 3060 12Gb VRAM, Windows 11 Pro, Comfy and models are all on SSDs. Thank you! :)

[mno1337]

I got it fixed. I updated torch to 2.5.0 then downgraded again to 2.4.1 and now my text encode takes like 6 seconds.

Den 7 sep. 2024 11:51, kI 11:51, Zdeto @.***> skrev:

4ltData.json comfyui.log

All right, I've attached the workflow and the comfyui log. The first run of the workflow took 444s, the next run just ~250s, because the cliptext was already encoded. This workflow has a Facedetailer in it, but the prompt I run generated an image without discernable faces, so this step was basically skipped. But I used to run this workflow (including the Facedetailer steps) in under 300s!

My hardware specs... i7-10700K, 32Gb RAM, RTX 3060 12Gb VRAM, Windows 11 Pro, Comfy and models are all on SSDs. Thank you! :)

-- Reply to this email directly or view it on GitHub: https://github.com/comfyanonymous/ComfyUI/issues/4745#issuecomment-2335136325 You are receiving this because you authored the thread.

Message ID: @.***>

[Zdeto]

I got it fixed. I updated torch to 2.5.0 then downgraded again to 2.4.1 and now my text encode takes like 6 seconds.

Ok, how exactly did you do that? Because I get an error when I try to update torch to 2.5.0 "ERROR: No matching distribution found for torch==2.5.0"

[mno1337]

https://pytorch.org/ Choose Preview (Nightly)

Den 7 sep. 2024 12:40, kI 12:40, Zdeto @.***> skrev:

I got it fixed. I updated torch to 2.5.0 then downgraded again to 2.4.1 and now my text encode takes like 6 seconds.

Ok, how exactly did you do that? Because I get an error when I try to update torch to 2.5.0 "ERROR: No matching distribution found for torch==2.5.0"

-- Reply to this email directly or view it on GitHub: https://github.com/comfyanonymous/ComfyUI/issues/4745#issuecomment-2335147843 You are receiving this because you authored the thread.

Message ID: @.***>

[Zdeto]

https://pytorch.org/ Choose Preview (Nightly) Den 7 sep. 2024 12:40, kI 12:40, Zdeto @.***> skrev: …

I got it fixed. I updated torch to 2.5.0 then downgraded again to 2.4.1 and now my text encode takes like 6 seconds. Ok, how exactly did you do that? Because I get an error when I try to update torch to 2.5.0 "ERROR: No matching distribution found for torch==2.5.0" -- Reply to this email directly or view it on GitHub: #4745 (comment) You are receiving this because you authored the thread. Message ID: @.***>

Please bare with me, I must be doing something wrong... So, I run this command and got errors about Requirement already satisfied.

e:\ComfyUI_windows_portable>python_embeded\python.exe -s -m pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu121 Looking in indexes: https://download.pytorch.org/whl/nightly/cu121 Requirement already satisfied: torch in e:\comfyui_windows_portable\python_embeded\lib\site-packages (2.4.0+cu121) Requirement already satisfied: torchvision in e:\comfyui_windows_portable\python_embeded\lib\site-packages (0.19.0+cu121) Requirement already satisfied: torchaudio in e:\comfyui_windows_portable\python_embeded\lib\site-packages (2.4.0+cu121) Requirement already satisfied: filelock in e:\comfyui_windows_portable\python_embeded\lib\site-packages (from torch) (3.15.4) Requirement already satisfied: typing-extensions>=4.8.0 in e:\comfyui_windows_portable\python_embeded\lib\site-packages (from torch) (4.12.2) Requirement already satisfied: sympy in e:\comfyui_windows_portable\python_embeded\lib\site-packages (from torch) (1.12.1) Requirement already satisfied: networkx in e:\comfyui_windows_portable\python_embeded\lib\site-packages (from torch) (3.3) Requirement already satisfied: jinja2 in e:\comfyui_windows_portable\python_embeded\lib\site-packages (from torch) (3.1.4) Requirement already satisfied: fsspec in e:\comfyui_windows_portable\python_embeded\lib\site-packages (from torch) (2024.5.0) Requirement already satisfied: numpy in e:\comfyui_windows_portable\python_embeded\lib\site-packages (from torchvision) (1.26.4) Requirement already satisfied: pillow!=8.3.*,>=5.3.0 in e:\comfyui_windows_portable\python_embeded\lib\site-packages (from torchvision) (10.4.0) Requirement already satisfied: MarkupSafe>=2.0 in e:\comfyui_windows_portable\python_embeded\lib\site-packages (from jinja2->torch) (2.1.5) Requirement already satisfied: mpmath<1.4.0,>=1.1.0 in e:\comfyui_windows_portable\python_embeded\lib\site-packages (from sympy->torch) (1.3.0)

[Zdeto]

All right, I had to uninstall first... stupid me :))) So, I uninstalled it python_embeded\python.exe -s -m pip uninstall torch torchvision torchaudio then run again python_embeded\python.exe -s -m pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu121

After that, I fired up Comfy... got some errors of course, but it started. Run the workflow and bam, ~3s for the CLIPTextEncodeFlux. But, as it happened to you, it doubled my sampler's s/it. Uninstalled again, then reinstalled torch 2.4.1. Ran the Comfy again, reload the workflow and now it taskes ~20s for the CLIPTextEncodeFlux and went back to the inference time. Still a bit long, but at least I got back to under 300s for this workflow to run :)

[mno1337]

Hehe yea same here, sampler got alot slower with 2.5.0!

Den 7 sep. 2024 16:34, kI 16:34, Zdeto @.***> skrev:

All right, I had to uninstall first... stupid me :))) So, I uninstalled it python_embeded\python.exe -s -m pip uninstall torch torchvision torchaudio then run again python_embeded\python.exe -s -m pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu121

After that, I fired up Comfy... got some errors of course, but it started. Run the workflow and bam, ~3s for the CLIPTextEncodeFlux. But, as it happened to you, it doubled my sampler's it/s. Uninstalled again, then reinstalled torch 2.4.1. Ran the Comfy again, reload the workflow and now it taskes ~20s for the CLIPTextEncodeFlux and went back to the inference time. Still a bit long, but at least I got back to under 300s for this workflow to run :)

-- Reply to this email directly or view it on GitHub: https://github.com/comfyanonymous/ComfyUI/issues/4745#issuecomment-2335271548 You are receiving this because you authored the thread.

Message ID: @.***>

[mno1337]

i updated comfy and now its damn slow again... 80-90s

[Zdeto]

Yeah, made the same mistake :(

[mno1337]

"#112 [CLIPTextEncodeFlux]: 143.94s".... Getting sick of it haha

Den 13 sep. 2024 16:13, kI 16:13, Zdeto @.***> skrev:

Yeah, made the same mistake :(

-- Reply to this email directly or view it on GitHub: https://github.com/comfyanonymous/ComfyUI/issues/4745#issuecomment-2349063085 You are receiving this because you authored the thread.

Message ID: @.***>

[Zdeto]

eventually I solved it as before, but it's annoying...

[mno1337]

Yea but it will come back ;)

Den 13 sep. 2024 20:02, kI 20:02, Zdeto @.***> skrev:

eventually I solved it as before, but it's annoying...

-- Reply to this email directly or view it on GitHub: https://github.com/comfyanonymous/ComfyUI/issues/4745#issuecomment-2349701214 You are receiving this because you authored the thread.

Message ID: @.***>

[Zdeto]

I can't believe we're the only lucky ones to encounter this. Nobody else has this issue?! :))

[mno1337]

Yea i know, its unbelievable... You should read about it everywhere! What hardware do you have?

Den 14 sep. 2024 10:25, kI 10:25, Zdeto @.***> skrev:

I can't believe we're the only lucky ones to encounter this. Nobody else has this issue?! :))

-- Reply to this email directly or view it on GitHub: https://github.com/comfyanonymous/ComfyUI/issues/4745#issuecomment-2350911297 You are receiving this because you authored the thread.

Message ID: @.***>

[Zdeto]

Look above, at the first part of this thread. I have listed my hardware and comfy.log (there are hardware specs in it)

[mno1337]

Look above, at the first part of this thread. I have listed my hardware and comfy.log (there are hardware specs in it)

Ah i see.. not same as me so nothing to do with that. Everyone i asked have never had problem with it so idk wtf is going on, annoying as hell anyway :D

[RaySteve312]

I have the problem too. it takes the same amount of time as the sampler does.

[ltdrdata]

For those experiencing this issue, please check the following:

1. Does the same phenomenon occur when using a normal clip model instead of gguf?
2. When checking memory usage through task manager, is swapping not occurring?

[swim2sun]

got same issue, and I found GPU is not using during CLIPTextEncodeFlux runing

[mno1337]

I found my problem. It was when i restarted comfy from the manager. Then it doesnt start with my args.

Den 11 okt. 2024 21:25, kI 21:25, Xiangyang You @.***> skrev:

got same issue, and I found GPU is not using during CLIPTextEncodeFlux runing

-- Reply to this email directly or view it on GitHub: https://github.com/comfyanonymous/ComfyUI/issues/4745#issuecomment-2408003863 You are receiving this because you authored the thread.

Message ID: @.***>

[swim2sun]

got same issue, and I found GPU is not using during CLIPTextEncodeFlux runing

I fixed by following steps, I have no idea which one works, pls try 3rd step firstly:

1. run update_comfyui_and_python_dependencies.bat
2. install pytorch 2.4.0 manually
3. remove the --lowvram in run_nvidia_gpu.bat

[ltdrdata]

I found my problem. It was when i restarted comfy from the manager. Then it doesnt start with my args. Den 11 okt. 2024 21:25, kI 21:25, Xiangyang You @.> skrev: … got same issue, and I found GPU is not using during CLIPTextEncodeFlux runing -- Reply to this email directly or view it on GitHub: #4745 (comment) You are receiving this because you authored the thread. Message ID: @.>

oh...... I need to check that. edit: As I tested it, when rebooting through the manager, all args are being passed correctly without any omissions.

[github-actions[bot]]

This issue is being marked stale because it has not had any activity for 30 days. Reply below within 7 days if your issue still isn't solved, and it will be left open. Otherwise, the issue will be closed automatically.