Skip to content

fix: Mixed Prec memory improvements and better default configs (converge-able) #32

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
Merged
merged 9 commits into from
Mar 25, 2025
Merged

fix: Mixed Prec memory improvements and better default configs (converge-able) #32

merged 9 commits into from
Mar 25, 2025

Conversation

SahilJain314
Copy link
Contributor

@SahilJain314 SahilJain314 commented Mar 23, 2025 •
edited
Loading

  • improved offloading by doing bf16 weight cast on IPC weights then offloading the full model
  • MP with FSDP MP + autocast MP
  • also improved startup speed by having vllm init with dummy weights then refitting.
  • changed default configs from "technically runs" to "will actually improve model"

#13

@SahilJain314
Multiprocessing memory improvements and better defaults
10c114f 
Signed-off-by: Sahil Jain <sahilj@nvidia.com>
@SahilJain314 SahilJain314 mentioned this pull request Mar 23, 2025
Closed
parthchadha
parthchadha previously approved these changes Mar 23, 2025
View reviewed changes
@SahilJain314 SahilJain314 changed the title Multiprocessing memory improvements and better default configs (converge-able) Mixed Prec memory improvements and better default configs (converge-able) Mar 23, 2025
@terrykong @SahilJain314
fix: ray.sub race condition when overlapping srun commands on same node
e89b13e 
trying a different approach

Signed-off-by: Terry Kong <terryk@nvidia.com>
parthchadha
parthchadha reviewed Mar 24, 2025
View reviewed changes
@SahilJain314
Fixed default LR 8B
774219d 
Signed-off-by: Sahil Jain <sahilj@nvidia.com>
@SahilJain314
lint fix
06c73ad 
Signed-off-by: Sahil Jain <sahilj@nvidia.com>
@parthchadha parthchadha changed the title Mixed Prec memory improvements and better default configs (converge-able) fix: Mixed Prec memory improvements and better default configs (converge-able) Mar 24, 2025
@SahilJain314
updated tests for dummy vllm init
79427e4 
Signed-off-by: Sahil Jain <sahilj@nvidia.com>
@SahilJain314 SahilJain314 merged commit bd7e4b0 into main Mar 25, 2025
13 checks passed
@SahilJain314 SahilJain314 deleted the sahilj/mp_fix branch March 25, 2025 07:08
KiddoZhu pushed a commit that referenced this pull request May 6, 2025
@SahilJain314 @terrykong @parthchadha
fix: Mixed Prec memory improvements and better default configs (conve…
02fe30d 
…rge-able) (#32)

Signed-off-by: Sahil Jain <sahilj@nvidia.com>
Signed-off-by: Terry Kong <terryk@nvidia.com>
Co-authored-by: Terry Kong <terryk@nvidia.com>
Co-authored-by: Parth Chadha <pchadha@nvidia.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@parthchadha parthchadha parthchadha 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.
3 participants
@SahilJain314 @parthchadha @terrykong