change of model download path?

[RYG81]

Hi, I don't know if true but download path keep changes for models. some time it downloads in custom node folder and sometime in cache folder under huggingface hub folder, and now we have folder for it in models folder. Any way to fix it to specific folder?

[gokayfem]

i made a new update to repo yesterday, now all of the files downloads into models/LLavacheckpoints, this is more convenient.

[RYG81]

So how can I move old downloads to this new location and save download data and time?

[gokayfem]

you can move them and it will work, even if you dont move them huggingface will find them i think. also becareful move only the files starts with 'filesfor...'

[RYG81]

Just moving this folder data to new folders in model folder

[gokayfem]

yes you can move the folder files_for_uform_gen2_qwen to models/LLavacheckpoints. Dont forget to update the repository.

[RYG81]

thanks for the guidance. Hope to have more vision model support added. All the best and keep good work.

[gokayfem]

thanks for the guidance. Hope to have more vision model support added. All the best and keep good work.

which one do you want

[RYG81]

Suggestion for vision model - https://huggingface.co/spaces/Vision-CAIR/MiniGPT-v2 also it keeps downloading models, i have created all nodes and downloaded all required models in folder but still everytime I run it downloads models under user .cache\huggingface\hub folder Please fix this as it takes very long and hard to break downloading process

Also, i suggest print download folder location so that we know its downloading on right path

[RYG81]

I am getting this error when using kosmos2 and internlm

Error occurred when executing Kosmos2model:

Allocation on device 0 would exceed allowed memory. (out of memory) Currently allocated : 9.50 GiB Requested : 508.10 MiB Device limit : 8.00 GiB Free (according to CUDA): 0 bytes PyTorch limit (set by user-supplied memory fraction) : 17179869184.00 GiB

File "D:\000AI\ComfyUI\ComfyUI\execution.py", line 149, in recursive_execute obj = class_def() ^^^^^^^^^^^ File "D:\000AI\ComfyUI\ComfyUI\custom_nodes\ComfyUI_VLM_nodes\nodes\kosmos2.py", line 55, in init self.predictor = KosmosModelPredictor() ^^^^^^^^^^^^^^^^^^^^^^ File "D:\000AI\ComfyUI\ComfyUI\custom_nodes\ComfyUI_VLM_nodes\nodes\kosmos2.py", line 21, in init self.model = AutoModelForVision2Seq.from_pretrained(self.model_path).to(self.device) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:\000AI\ComfyUI\python_embeded\Lib\site-packages\transformers\modeling_utils.py", line 2595, in to return super().to(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:\000AI\ComfyUI\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1152, in to return self._apply(convert) ^^^^^^^^^^^^^^^^^^^^ File "D:\000AI\ComfyUI\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 802, in _apply module._apply(fn) File "D:\000AI\ComfyUI\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 802, in _apply module._apply(fn) File "D:\000AI\ComfyUI\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 802, in _apply module._apply(fn) File "D:\000AI\ComfyUI\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 825, in _apply param_applied = fn(param) ^^^^^^^^^ File "D:\000AI\ComfyUI\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1150, in convert return t.to(device, dtype if t.is_floating_point() or t.is_complex() else None, non_blocking) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

[gokayfem]

internlm is a heavy model even it's quantized version. also its not downloading, its fetching your old files from this repository from the root.

[RYG81]

https://github.com/Sanster/VLM-demos this repo lists many VLMs, if there is a way to integrate all of them in comfy.

[RYG81]

Models keep downloading, i have used ufromgen2 before and have all models on location, but still it keeps downloading

https://github.com/gokayfem/ComfyUI_VLM_nodes/assets/10570236/c99498e1-b27c-4abc-b4cd-b33c331e044d