oneflow

oneflow copied to clipboard

Dev alphafold fused attn

1

fused kernels in alphafold

ofhwei

This reverts commit 1ba7158d47b5fbb4474a2f3e3ccb9427ccd7e2b5.

howin98

重构ONEFLOW_VM_COMPUTE_ON_WORKER_THREAD，使其默认值为false，同时只对默认的worker thread 0生效。

lixinqi

graph support flow.autograd.grad

8

hjchen2

## Summary ```python import torch import oneflow def dummy_sample_deter(mod): num_elems = 20 sample = mod.arange(num_elems) sample = sample / num_elems print(mod.__name__, sample.dtype) dummy_sample_deter(oneflow) dummy_sample_deter(torch) ``` ```bash loaded library: /lib/x86_64-linux-gnu/libibverbs.so.1 oneflow...

jackalcooper

## Summary When importing oneflow as torch, function `allclose` not avaliable. ## Code to reproduce bug Please post a minimal example to repro the bug. GitHub Gist or repo is...

jackalcooper

关于oneflow一份代码内使用连续使用两个不同模型出现的问题

4

What is your OneFlow installation (pip, source, dockerhub): pip OS:Linux version 5.8.0-63-generic (buildd@lgw01-amd64-035) (gcc (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0, GNU ld (GNU Binutils for Ubuntu) 2.34) OneFlow version (run python3 -m oneflow...

yaoxinthu

liujuncheng

fp16 division is inconsistent with torch

1

## Summary fp16除法操作返回结果类型和torch不一致 ## Code to reproduce bug ```python >>>import oneflow as flow >>>a = flow.randn(3, 3, dtype=flow.float16).cuda() >>>b = flow.randn(3, 3, dtype=flow.float16).cuda() >>>a/b tensor([[-2.1495e-03, 1.5983e+00, -5.2973e-01], [-1.7968e-01, -4.0361e+00, 5.4459e-01],...

ofhwei

Add a switch for memory share strategy

2

Add a switch for the memory share strategy. It would be off by default. It solves this issue https://github.com/Oneflow-Inc/oneflow/issues/9508

Yipeng1994