Skip to content

[BUG] Load fused_adam_cuda failed #2874

Description

@Fazziekey

Describe the bug
I install the deepspeed with DS_BUILD_OPS=1 pip install deepspeed and compile the extra keenal, but I get err when I use zero optimizer, I get fused_adam_cuda = FusedAdamBuilder().load(), TypeError: 'NoneType' object is not callable

image

Expected behavior
A clear and concise description of what you expected to happen.

ds_report output
I run ds_report and get this error
image

System info (please complete the following information):

  • OS: CentOS
  • GPU count and types: 1machines with x8 A100s each
  • Python version: 3.9.12
  • Cuda:11.3
  • torch:1.13

Launcher context
deepspeed --num_gpus=2 train.py

Metadata

Metadata

Assignees

Labels

bugSomething isn't workingbuildImprovements to the build and testing systems.

Type

No type

Fields

No fields configured for issues without a type.

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions