Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Fix Low CPU Memory Mode Issues for Quantized Peft #90

Merged
merged 8 commits into from
Oct 17, 2024
Merged

Fix Low CPU Memory Mode Issues for Quantized Peft #90

merged 8 commits into from
Oct 17, 2024

Conversation

fabianlim
Copy link
Contributor

@fabianlim fabianlim commented Oct 11, 2024 •
edited
Loading

Since transformers >=0.45 we are having issues with low_cpu_mem_mode as described in #83 . This PR address them

Note: this PR fixes will apply only if the following commits are applied:

Regressionss

Mistral 7B

image image
image image

Granite 3b

image image
image image

@fabianlim fabianlim marked this pull request as draft October 11, 2024 03:56
@fabianlim fabianlim linked an issue Oct 14, 2024 that may be closed by this pull request
Distributed Training Problems for QLoRA models with Transformers pre-release 4.45 #83
Closed
@fabianlim
address issue 2 in #83
aacd5bd 
Signed-off-by: Yu Chin Fabian Lim <[email protected]>
@fabianlim
properly handle broadcast of adapters
1e99ee9 
Signed-off-by: Yu Chin Fabian Lim <[email protected]>
@fabianlim
handle param_init_fn_tied_param
0f5c9ae 
Signed-off-by: Yu Chin Fabian Lim <[email protected]>
@fabianlim
trl version error
18aa160 
Signed-off-by: Yu Chin Fabian Lim <[email protected]>
@fabianlim
tied weights fix and meta fix for autogptq
e02993d 
Signed-off-by: Yu Chin Fabian Lim <[email protected]>
@fabianlim
update readme
510e0d1 
Signed-off-by: Yu Chin Fabian Lim <[email protected]>
@fabianlim
fmt + lint
a85faac 
Signed-off-by: Yu Chin Fabian Lim <[email protected]>
@fabianlim fabianlim marked this pull request as ready for review October 14, 2024 06:25
@fabianlim
upgrade granite benches
5a24f02 
Signed-off-by: Yu Chin Fabian Lim <[email protected]>
anhuong
anhuong reviewed Oct 16, 2024
View reviewed changes
Copy link
Collaborator

@anhuong anhuong left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Left a few questions for my understanding, I don't know if I'm the best person to review this change but overall if the changes were tested and it successfully loaded the model for GPTQ with low_cpu_mem_usage set then that seems good to me

plugins/accelerated-peft/src/fms_acceleration_peft/gptqmodel/models/base.py Show resolved Hide resolved
plugins/fused-ops-and-kernels/src/fms_acceleration_foak/framework_plugin_fast_quantized_peft.py Show resolved Hide resolved
plugins/accelerated-peft/src/fms_acceleration_peft/framework_plugin_autogptq.py Show resolved Hide resolved
anhuong
anhuong approved these changes Oct 17, 2024
View reviewed changes
Copy link
Collaborator

@anhuong anhuong left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for answering my questions, this looks good to me but might want someone else to give it a closer review

@fabianlim fabianlim merged commit fc78b55 into main Oct 17, 2024
6 checks passed
@fabianlim fabianlim deleted the fix/quant branch October 17, 2024 23:12
@fabianlim fabianlim mentioned this pull request Oct 25, 2024
Merged
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@anhuong anhuong anhuong approved these changes

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

Distributed Training Problems for QLoRA models with Transformers pre-release 4.45
2 participants
@fabianlim @anhuong