-
Hi,
version 2.1.2 has been compiled with a wrong java version, so it leads to a crash at startup.
[12:04:26] [main/ERROR]: Minecraft has crashed!
net.fabricmc.loader.impl.FormattedException: ja…
-
## 🚀 Feature
Improved Transformer and MultiHeadAttention design
## Motivation
Current Transformer and MultiHeadAttention design is not a very `PyTorch`-ish. The `in_proj` is not even a `nn.Mo…
-
A new design for the home-screen was developed, which does not require horizontal scrolling:
https://www.figma.com/file/vUnfD52FdhTmXq4Wjd5ieI/Bmm-Web-2023?type=design&node-id=1004-5327&mode=design&t=…
-
Thanks a lot for your great work!
I'm trying using flash attn to replace nv's megatron origin multi head attention, but while i'm trying to run it end to end & observe its loss convergence, it diff…
-
## 🐛 Bug
The high rank tensor contraction with einsum `"jk,ijkl->il"` is about 16x slower than numpy.
## To Reproduce
import time
import torch
import numpy as np
nb = 160…
-
Hello @ahojnnes
Is it possible to orient/transform image (in image coordinate system) according to reference image whose colmap poses: R, t, projection matrix, FoVx, FoVy, image_width, image_height …
-
-
### 🐛 Describe the bug
# RuntimeError: Double and complex datatype matmul is not supported in oneDNN
"test_comprehensive_nn_functional_linear_xpu_bfloat16",
"test_comprehensive_nn_funct…
-
Thanks for the nice work. When I tried **mustard0**, and it works fine. But when I try to run other objects such as driller (example in your link) and cheezit (collected by myself), it had this error:…
-
if not self.cfg.aux_loss:
pred = torch.bmm(query_output, pixel_output.flatten(2))
pred = rearrange(pred, 'b l (h w) -> b l h w', h=h, w=w)
else:
for l…