xformers

xformers copied to clipboard

xFormes没有加载CUDA，无法使用CUDA

5

在运行CUDA时报告: WARNING[XFORMERS]: xFormers can't load C++/CUDA extensions. xFormers was built for: PyTorch 2.1.0+cu118 with CUDA 1106 (you have 2.0.1+cu118) Python 3.9.16 (you have 3.10.12) Please reinstall xformers (see https://github.com/facebookresearch/xformers#installing-xformers) Memory-efficient...

xiahongyu939

Problems installing xFormers in Kaggle notebook

3

# ❓ Questions and Help I've been trying to install xFormers in a auxilliary Kaggle notebook (in the /kaggle/working folder) which will be used as a utility script in another...

probit2011

Torch.onnx.export breaks when memory_efficient_attention

4

# 🐛 Bug RuntimeError: unsupported output type: int, from operator: xformers::efficient_attention_forward_cutlass ## Command import torch import xformers.ops as xops import torch.nn as nn class Attention(nn.Module): def forward(self, q,k,v): x =...

deyiluobo

memory_efficient_attention fw produce inconsistent results

1

# ❓ Questions and Help memory_efficient_attention fw produce inconsistent results not sure what was going on? incorrect built? some specific versions combinations? for some combinations: xformers torch CUDA GPU CUDA...

ShijunK

update get_split_k to fix a performance regression on FA decode

4

## What does this PR do? Fixes a performance regress for FA decode ## Before submitting - [x] Did you have fun? - [x] Did you read the [contributor guideline](https://github.com/facebookresearch/xformers/blob/master/CONTRIBUTING.md)?...

scxiao

The original code directly disabled the support for fused swiglu, but pytorch2.0.0 has fixed the issue in autocast of bf16.

3

A small PR. This PR addresses the issue of disabling fused swiglu in swiglu_op.py. The original code directly disabled the support for fused swiglu, with the comment "Let's disable autocast...

Gstdioh

# 🐛 Bug ## Command ## To Reproduce Steps to reproduce the behavior: 1. 2. 3. ## Expected behavior ## Environment Please copy and paste the output from the environment...

Sariel2

[Feature Request] Sparse tensor support for `attn_bias` in `memory_efficient_attention`

1

# 🚀 Feature I think that it would be very helpful to allow sparse tensors to be used as masks (`attn_bias`) in the `memory_efficient_attention` function. ## Motivation As the documentation...

davidbuterez