was-node-suite-comfyui
was-node-suite-comfyui copied to clipboard
WASasquatch
could you please update the blip required version of transformer to version 36 or higher
That is the last version of Transformers that Transformers BLIP code works on, which is why it's pinned. A lot of people still use BLIP, and most can't run BLIP2.
How to use transformers==4.26.0 in a separate venv for WAS Node Suite nodes (where can I get a step-by-step tutorial for configuring venv for a separate node in Comfi)? transformers==4.36.0 is needed for ClipInterrogator node (in Mixlab nodes package) to work correctly :``(
这是 Transformers BLIP 代码工作的最后一个 Transformers 版本，这就是它被固定的原因。很多人仍然使用 BLIP，大多数人无法运行 BLIP2。
https://github.com/WASasquatch/was-node-suite-comfyui/issues/369 There is a conflict between the current locked Transformer version and the latest d14bdb18 version of ComfyUI
That is the last version of Transformers that Transformers BLIP code works on, which is why it's pinned. A lot of people still use BLIP, and most can't run BLIP2.
I think it's better to upgrade transformers. Most nodes require higher version of transformers. You can take a look at https://github.com/sipherxyz/comfyui-art-venture. It's blip caption node works well with transformers==4.39.1. It works with the model automaticly downloaded by your blip node.
The code I am using is not compatible. If you know how to PR an upgrade of the code to use high transformers that would be fine. But I have no plans to upgrade WAS-NS at this time. As noted on the readme it's unmanaged at this time. PRs are welcome though.
The code I am using is not compatible. If you know how to PR an upgrade of the code to use high transformers that would be fine. But I have no plans to upgrade WAS-NS at this time. As noted on the readme it's unmanaged at this time. PRs are welcome though.
Got it. Thanks for reply.
https://github.com/WASasquatch/was-node-suite-comfyui/pull/393
Refactored BLIP!
FYI the reason Art Venture's BLIP works is because it's literally the original BLIP code copy-pasted. It's not transformers based.
#393
Refactored BLIP!
FYI the reason Art Venture's BLIP works is because it's literally the original BLIP code copy-pasted. It's not transformers based.
Thanks a lot for update!
When i searching model "blip-vqa-base" on hg, i found a bf16 version of this model,half memory size of the original one. Test several images, it seems return the same result.
- https://hf-mirror.com/gospacedev/blip-image-captioning-base-bf16
I designed it so you can use any model you want that works with transformers.
Is the b16 also faster?