Skip to content

Conversation

weixiao-huang
Copy link
Contributor

@weixiao-huang weixiao-huang commented Sep 5, 2025

Purpose

Resolve #24163

Test Plan

cd examples/offline_inference
python3 rlhf_colocate.py

Test Result

rlhf_colocate.py is executed successfully

image

@mergify mergify bot added documentation Improvements or additions to documentation v1 labels Sep 5, 2025
Copy link
Contributor

@gemini-code-assist gemini-code-assist bot left a comment

Choose a reason for hiding this comment

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

Code Review

This pull request introduces a new mechanism for fast, in-place model weight updates using ZMQ for inter-process communication, which is a good improvement for RLHF-style workloads. The implementation on the worker side is mostly solid, but it relies on a brittle internal detail of PyTorch. Additionally, the example script demonstrating this feature contains a performance bottleneck that undermines the goal of "fast" updates. My review provides suggestions to address these issues.

@weixiao-huang weixiao-huang force-pushed the feat/fast-inplace-model-update branch from 05a7b7d to 198631d Compare September 5, 2025 06:59
Copy link

mergify bot commented Sep 6, 2025

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @weixiao-huang.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

@mergify mergify bot added the needs-rebase label Sep 6, 2025
@22quinn 22quinn added the rl Related to RL workflows label Sep 7, 2025
@mergify mergify bot removed the needs-rebase label Sep 8, 2025
@weixiao-huang weixiao-huang force-pushed the feat/fast-inplace-model-update branch from e63d2e7 to 348093a Compare September 8, 2025 08:45
@weixiao-huang weixiao-huang force-pushed the feat/fast-inplace-model-update branch from 348093a to a64f369 Compare September 8, 2025 12:09
Signed-off-by: huangweixiao <huangweixiao@msh.team>
Signed-off-by: huangweixiao <huangweixiao@msh.team>
@weixiao-huang weixiao-huang force-pushed the feat/fast-inplace-model-update branch 2 times, most recently from 1cfe7fb to b73dc08 Compare September 9, 2025 05:23
Signed-off-by: huangweixiao <huangweixiao@msh.team>
@weixiao-huang weixiao-huang force-pushed the feat/fast-inplace-model-update branch from b73dc08 to 35e57e5 Compare September 9, 2025 07:03
Signed-off-by: youkaichao <youkaichao@gmail.com>
Signed-off-by: youkaichao <youkaichao@gmail.com>
@youkaichao youkaichao changed the title [feat] fast inplace model update [RL] fast weight update with zmq + ipc handles Sep 9, 2025
Copy link
Member

@youkaichao youkaichao left a comment

Choose a reason for hiding this comment

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

Great work!

@youkaichao youkaichao enabled auto-merge (squash) September 9, 2025 08:12
@github-actions github-actions bot added the ready ONLY add when PR is ready to merge/full CI is needed label Sep 9, 2025
@youkaichao youkaichao disabled auto-merge September 9, 2025 08:57
@youkaichao youkaichao merged commit 3d2a2de into vllm-project:main Sep 9, 2025
19 of 26 checks passed
eicherseiji pushed a commit to eicherseiji/vllm that referenced this pull request Sep 9, 2025
Signed-off-by: huangweixiao <huangweixiao@msh.team>
Signed-off-by: youkaichao <youkaichao@gmail.com>
Co-authored-by: youkaichao <youkaichao@gmail.com>
skyloevil pushed a commit to skyloevil/vllm that referenced this pull request Sep 13, 2025
Signed-off-by: huangweixiao <huangweixiao@msh.team>
Signed-off-by: youkaichao <youkaichao@gmail.com>
Co-authored-by: youkaichao <youkaichao@gmail.com>
FeiDaLI pushed a commit to FeiDaLI/vllm that referenced this pull request Sep 25, 2025
Signed-off-by: huangweixiao <huangweixiao@msh.team>
Signed-off-by: youkaichao <youkaichao@gmail.com>
Co-authored-by: youkaichao <youkaichao@gmail.com>
xuebwang-amd pushed a commit to xuebwang-amd/vllm that referenced this pull request Oct 10, 2025
Signed-off-by: huangweixiao <huangweixiao@msh.team>
Signed-off-by: youkaichao <youkaichao@gmail.com>
Co-authored-by: youkaichao <youkaichao@gmail.com>
Signed-off-by: xuebwang-amd <xuebwang@amd.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

documentation Improvements or additions to documentation ready ONLY add when PR is ready to merge/full CI is needed rl Related to RL workflows v1

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[RFC]: Support fast inplace model update by shared IPC buffer

3 participants