[Misc]: Implement CPU/GPU swapping in BlockManagerV2 #3834

Kaiyang-Chen · 2024-04-03T22:16:38Z

A draft PR for #3666, supporting CPU-GPU swapping for Block Manager V2.

PR Checklist (Click to Expand)

Thank you for your contribution to vLLM! Before submitting the pull request, please ensure the PR meets the following criteria. This helps vLLM maintain the code quality and improve the efficiency of the review process.

PR Title and Classification

Only specific types of PRs will be reviewed. The PR title is prefixed appropriately to indicate the type of change. Please use one of the following:

[Bugfix] for bug fixes.
[CI/Build] for build or continuous integration improvements.
[Doc] for documentation fixes and improvements.
[Model] for adding a new model or improving an existing model. Model name should appear in the title.
[Frontend] For changes on the vLLM frontend (e.g., OpenAI API server, LLM class, etc.)
[Kernel] for changes affecting CUDA kernels or other compute kernels.
[Core] for changes in the core vLLM logic (e.g., LLMEngine, AsyncLLMEngine, Scheduler, etc.)
[Hardware][Vendor] for hardware-specific changes. Vendor name should appear in the prefix (e.g., [Hardware][AMD]).
[Misc] for PRs that do not fit the above categories. Please use this sparingly.

Note: If the PR spans more than one category, please include all relevant prefixes.

Code Quality

The PR need to meet the following code quality standards:

We adhere to Google Python style guide and Google C++ style guide.
Pass all linter checks. Please use format.sh to format your code.
The code need to be well-documented to ensure future contributors can easily understand the code.
Include sufficient tests to ensure the project to stay correct and robust. This includes both unit tests and integration tests.
Please add documentation to docs/source/ if the PR modifies the user-facing behaviors of vLLM. It helps vLLM user understand and utilize the new features or changes.

Notes for Large Changes

Please keep the changes as concise as possible. For major architectural changes (>500 LOC excluding kernel/data/config/test), we would expect a GitHub issue (RFC) discussing the technical design and justification. Otherwise, we will tag it with rfc-required and might not go through the PR.

What to Expect for the Reviews

The goal of the vLLM team is to be a transparent reviewing machine. We would like to make the review process transparent and efficient and make sure no contributor feel confused or frustrated. However, the vLLM team is small, so we need to prioritize some PRs over others. Here is what you can expect from the review process:

After the PR is submitted, the PR will be assigned to a reviewer. Every reviewer will pick up the PRs based on their expertise and availability.
After the PR is assigned, the reviewer will provide status update every 2-3 days. If the PR is not reviewed within 7 days, please feel free to ping the reviewer or the vLLM team.
After the review, the reviewer will put an action-required label on the PR if there are changes required. The contributor should address the comments and ping the reviewer to re-review the PR.
Please respond to all comments within a reasonable time frame. If a comment isn't clear or you disagree with a suggestion, feel free to ask for clarification or discuss the suggestion.

Thank You

Finally, thank you for taking the time to read these guidelines and for your interest in contributing to vLLM. Your contributions make vLLM a great tool for everyone!

cadedaniel

Thanks for working on this. I'm leaving some early feedback.

vllm/core/block_manager_v2.py

vllm/core/block/block_table.py

vllm/core/block_manager_v2.py

vllm/core/block/cpu_gpu_block_allocator.py

…an_swap_out

cadedaniel · 2024-04-08T19:23:44Z

Will take another pass tomorrow

cadedaniel · 2024-04-10T22:40:19Z

I passed through the code. I think this PR is mixing the abstraction boundaries too much. I wrote out some pseudocode for a better way of using the existing APIs -- take a look.

https://docs.google.com/document/d/157VCEOs5EIB1aztD51ZrT2RvdjxMx1spazyBM4LS4aQ/edit

Kaiyang-Chen · 2024-04-17T17:45:56Z

Hi Cade, It's ready for review. @cadedaniel

cadedaniel · 2024-04-17T18:31:27Z

OK, will take a look thurs or fri this week

cadedaniel

Looking a lot better! Thanks for improving the PR.

Some missing things:

We'll need e2e tests which verify swapping works and produces correct results. You can basically copy these tests and simply enable preemption by swap instead of by compute.
- vllm/tests/core/block/e2e/test_correctness.py
  
  Line 187 in c1b4e41
  
  def test_lookahead_greedy_equality_with_preemption(baseline_llm_generator,
- vllm/tests/core/block/e2e/test_correctness.py
  
  Line 99 in c1b4e41
  
  def test_v1_v2_greedy_equality_with_cow(baseline_llm_generator,
- vllm/tests/core/block/e2e/test_correctness.py
  
  Line 28 in c1b4e41
  
  def test_v1_v2_greedy_equality_with_preemption(baseline_llm_generator,
The unit test in this PR is pretty narrow in its coverage. I suggest writing the test at a lower level (at the BlockAllocator level), so that you can more easily test num_lookhead=[0, 2] and type=[prefix, naive].

vllm/core/block_manager_v2.py

vllm/core/block/prefix_caching_block.py

vllm/core/block/cpu_gpu_block_allocator.py

vllm/core/block_manager_v2.py

vllm/core/block/block_table.py

tests/core/block/test_block_manager_v2.py

cadedaniel

Is it ready for review again?

Kaiyang-Chen · 2024-04-30T18:36:21Z

Is it ready for review again?

Nope, still working on the test, will push the new version later today or tmr and notify you. Thanks for the attention!

Kaiyang-Chen · 2024-05-01T08:32:10Z

@cadedaniel Hi cade, i think it's ready for another round of review. One small point is currently the preemption_mode parameter in _preempt() are default to RECOMPUTE and never be passed in. Add some tiny changes around that to support testing preemption by swapping, should we add those in an separate MR?

Kaiyang-Chen · 2024-05-11T01:05:58Z

retrying

Hi Cade, can you help to retry the last failed test one more time? Thanks!

cadedaniel · 2024-05-11T01:30:47Z

retried

Kaiyang-Chen · 2024-05-11T02:28:35Z

retried

@cadedaniel TY! I think all tests are passed now.

Kaiyang-Chen · 2024-05-16T14:46:25Z

@cadedaniel Hi cade, solved the new conflicts. Can you help to retry and get it merged? TY!

cadedaniel · 2024-05-20T18:24:23Z

thanks @Kaiyang-Chen . i was out last week. I will get this merged this week, thanks for all your work!

Kaiyang-Chen · 2024-05-28T13:32:47Z

@cadedaniel resolve the conflicts again. Can we merge this now?

cadedaniel · 2024-05-30T00:12:32Z

Enabled auto merge

Kaiyang-Chen · 2024-05-31T01:16:23Z

Enabled auto merge

@cadedaniel It seems this cannot be auto merged because I don't have write access to the repo?

cadedaniel · 2024-06-03T20:37:23Z

Done! thanks again @Kaiyang-Chen

Kaiyang-Chen added 2 commits April 3, 2024 22:07

feat: support swap in/out for block manager v2

182d4a8

fix: linter

b6b4b8f

cadedaniel reviewed Apr 4, 2024

View reviewed changes

vllm/core/block_manager_v2.py Outdated Show resolved Hide resolved

vllm/core/block/block_table.py Outdated Show resolved Hide resolved

vllm/core/block_manager_v2.py Outdated Show resolved Hide resolved

Kaiyang-Chen added 2 commits April 4, 2024 02:44

fix: fix some bugs and add test

938d10e

fix: address comment

9181552

cadedaniel reviewed Apr 4, 2024

View reviewed changes

vllm/core/block_manager_v2.py Outdated Show resolved Hide resolved

vllm/core/block_manager_v2.py Outdated Show resolved Hide resolved

vllm/core/block_manager_v2.py Outdated Show resolved Hide resolved

vllm/core/block/cpu_gpu_block_allocator.py Outdated Show resolved Hide resolved

Kaiyang-Chen added 4 commits April 5, 2024 00:33

fix: reduce overestimate for can_swap_in

e9a907f

fix: reuse similar logic in can_swap_in to reduce overestimation in c…

dcff0e1

…an_swap_out

fix: refactor swap in/out logic

205dda1

misc: remove useless code

3bb125c

Kaiyang-Chen requested a review from cadedaniel April 5, 2024 09:09

Kaiyang-Chen added 6 commits April 12, 2024 02:40

fix: refactor can_swap_in/out

403a9bd

fix: remove unused code

3237d63

fix: remove unused code

4131247

fix: refactor swap in/out oprations

0067ddf

fix

b8aee85

fix

cba0f62

cadedaniel reviewed Apr 23, 2024

View reviewed changes

cadedaniel reviewed Apr 30, 2024

View reviewed changes

Kaiyang-Chen added 3 commits April 30, 2024 18:53

doc: adding docstring

0430758

test: adding e2e correstness test for preemption by swapping

fbb3099

fix

66a7bbd

remove import for __future__.annotations

35d391e

Kaiyang-Chen requested a review from cadedaniel May 1, 2024 08:35

Kaiyang-Chen and others added 3 commits May 13, 2024 22:03

Merge branch 'main' into dev_block_manager_v2_swap

228950a

merge

216eb76

Merge branch 'main' into dev_block_manager_v2_swap

e318e7e

Kaiyang-Chen and others added 2 commits May 28, 2024 21:17

Merge branch 'main' into dev_block_manager_v2_swap

4e1c511

fix: ci

862a5d4

cadedaniel enabled auto-merge (squash) May 30, 2024 00:12

cadedaniel approved these changes May 30, 2024

View reviewed changes

Kaiyang-Chen added 2 commits May 30, 2024 22:17

Merge branch 'main' into dev_block_manager_v2_swap

cb28e8f

fix: merge

29df092

auto-merge was automatically disabled May 30, 2024 14:19
Head branch was pushed to by a user without write access

cadedaniel merged commit 10c38e3 into vllm-project:main Jun 3, 2024
64 checks passed

cadedaniel mentioned this pull request Jun 3, 2024

[Misc]: Implement CPU/GPU swapping in BlockManagerV2 #3666

Closed

blinkbear pushed a commit to blinkbear/vllm that referenced this pull request Jun 6, 2024

[Misc]: Implement CPU/GPU swapping in BlockManagerV2 (vllm-project#3834)

13e8f41

robertgshaw2-neuralmagic pushed a commit to neuralmagic/nm-vllm that referenced this pull request Jun 11, 2024

[Misc]: Implement CPU/GPU swapping in BlockManagerV2 (vllm-project#3834)

a6f0725

joerunde pushed a commit to joerunde/vllm that referenced this pull request Jun 17, 2024

[Misc]: Implement CPU/GPU swapping in BlockManagerV2 (vllm-project#3834)

b0563b0

xjpang pushed a commit to xjpang/vllm that referenced this pull request Jun 27, 2024

[Misc]: Implement CPU/GPU swapping in BlockManagerV2 (vllm-project#3834)

66abf55

xjpang pushed a commit to xjpang/vllm that referenced this pull request Jul 8, 2024

[Misc]: Implement CPU/GPU swapping in BlockManagerV2 (vllm-project#3834)

77ee4d8

xjpang pushed a commit to xjpang/vllm that referenced this pull request Jul 24, 2024

[Misc]: Implement CPU/GPU swapping in BlockManagerV2 (vllm-project#3834)

e47501c

Temirulan pushed a commit to Temirulan/vllm-whisper that referenced this pull request Sep 6, 2024

[Misc]: Implement CPU/GPU swapping in BlockManagerV2 (vllm-project#3834)

b4e66a2

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Misc]: Implement CPU/GPU swapping in BlockManagerV2 #3834

[Misc]: Implement CPU/GPU swapping in BlockManagerV2 #3834

Kaiyang-Chen commented Apr 3, 2024 •

edited

Loading

cadedaniel left a comment

cadedaniel commented Apr 8, 2024

cadedaniel commented Apr 10, 2024

Kaiyang-Chen commented Apr 17, 2024

cadedaniel commented Apr 17, 2024

cadedaniel left a comment

cadedaniel left a comment

Kaiyang-Chen commented Apr 30, 2024

Kaiyang-Chen commented May 1, 2024

Kaiyang-Chen commented May 11, 2024

cadedaniel commented May 11, 2024

Kaiyang-Chen commented May 11, 2024

Kaiyang-Chen commented May 16, 2024

cadedaniel commented May 20, 2024

Kaiyang-Chen commented May 28, 2024

cadedaniel commented May 30, 2024

Kaiyang-Chen commented May 31, 2024

cadedaniel commented Jun 3, 2024

[Misc]: Implement CPU/GPU swapping in BlockManagerV2 #3834

[Misc]: Implement CPU/GPU swapping in BlockManagerV2 #3834

Conversation

Kaiyang-Chen commented Apr 3, 2024 • edited Loading

PR Title and Classification

Code Quality

Notes for Large Changes

What to Expect for the Reviews

Thank You

cadedaniel left a comment

Choose a reason for hiding this comment

cadedaniel commented Apr 8, 2024

cadedaniel commented Apr 10, 2024

Kaiyang-Chen commented Apr 17, 2024

cadedaniel commented Apr 17, 2024

cadedaniel left a comment

Choose a reason for hiding this comment

cadedaniel left a comment

Choose a reason for hiding this comment

Kaiyang-Chen commented Apr 30, 2024

Kaiyang-Chen commented May 1, 2024

Kaiyang-Chen commented May 11, 2024

cadedaniel commented May 11, 2024

Kaiyang-Chen commented May 11, 2024

Kaiyang-Chen commented May 16, 2024

cadedaniel commented May 20, 2024

Kaiyang-Chen commented May 28, 2024

cadedaniel commented May 30, 2024

Kaiyang-Chen commented May 31, 2024

cadedaniel commented Jun 3, 2024

Kaiyang-Chen commented Apr 3, 2024 •

edited

Loading