-
it should be
new_shape = tf.concat([tf.shape(input=q)[0:-1],[v.shape[-1]]],0)
instead of
new_shape=tf.concat([tf.shape(input=q)[0:-1],[v.shape[-1].value]],0)
-
Thank you for your code. How can I visualize prompt with textual inversion tokens?
-
There is a new promising model.
https://huggingface.co/docs/diffusers/main/en/api/pipelines/flux
It would be good to take this model pipeline from diffusers package and add a canvas and attention …
-
Bonjour,
Super cours, mais sur certains ordinateurs des utilisateurs peuvent avoir des problèmes pour importer la base, je me suis référé à ceci: https://tecadmin.net/resolved-unknown-collation-utf8m…
-
Hi there,
I wanted to understand more about part attention block about how does it actually attends to the parts of the object in the image. Moreover, I could not spot the specific code block written…
-
Dear professor, I have read your article recently and am very interested in the method of image reconstruction adopted in the article "Wavefront coding image reconstruction via physical prior and freq…
-
### Is there an existing issue for this?
- [X] I have searched the existing issues and checked the recent builds/commits
### What would your feature do ?
HuggingFace [recommends](https://huggingfac…
-
### Search before asking
- [X] I have searched the YOLOv5 [issues](https://github.com/ultralytics/yolov5/issues) and found no similar feature requests.
### Description
Hi,
I really appreciate thi…
-
### Attention | توجه
- [X] I searched and no similar issues were found | جستجو کردم و هیچ گزارش مشابهی پیدا نشد
### What Happened? | چه اتفاقی افتاده؟
error :
error: missing package metada…
-
The speed of flash-attention(QKV/softmax fused part) in megatron-LM is much slower than the native implementation of megatron-LM. What is the reason?
I used the latest version:https://github.com/…