Skip to content

Conversation

@stas00
Copy link
Collaborator

@stas00 stas00 commented Oct 20, 2021

In the memory doc this PR fixes broken formatting and adds an important note from @samyam to why fp16 gradients consume 4 bytes instead of 2.

@tjruwase

@tjruwase
Copy link
Contributor

Squashing before CI unit tests complete since this is a docs PR.

@tjruwase tjruwase merged commit dd22428 into deepspeedai:master Oct 20, 2021
@stas00 stas00 deleted the doc-typo branch October 20, 2021 22:19
@stas00
Copy link
Collaborator Author

stas00 commented Oct 22, 2021

@tjruwase, do you need to manually update https://deepspeed.readthedocs.io/en/stable/memory.html
it still shows the old borked version. Thank you so much!

@tjruwase
Copy link
Contributor

No, it is automatically updated. Perhaps give it a bit of time.

@tjruwase
Copy link
Contributor

Actually, it has been two days right? Hmm, this means something might be wrong. I will take a closer look.

@stas00
Copy link
Collaborator Author

stas00 commented Oct 22, 2021

it was merged 2 days ago, if it hasn't automatically updated it's unlikely it would ;)

@stas00
Copy link
Collaborator Author

stas00 commented Oct 22, 2021

Also I noticed it needs to be further updated to use the latest names of the offload config as it currently uses the deprecated cpu_offload, cpu_offload_params

@tjruwase
Copy link
Contributor

I will add to my TODO.

@stas00
Copy link
Collaborator Author

stas00 commented Oct 22, 2021

You're the best, @tjruwase!

Dipet pushed a commit to Dipet/DeepSpeed that referenced this pull request Oct 26, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants