[Bugfix][fast] Fix the get_num_blocks_touched logic #6849
Conversation
|
👋 Hi! Thank you for contributing to the vLLM project. Once the PR is approved and ready to go, please make sure to run full CI as it is required to merge (or just use auto-merge). To run full CI, you can do one of these:
🚀 |
|
@Kaiyang-Chen @cadedaniel pls review when you have time. |
|
Thanks for the PR. Can you help me understand the impact here of this bug? e.g. what kind of incorrect swapping behavior / corrupted output. Can we also add a unit test for the changed logic if possible? the block manager test is very high level unfortunately. cc @Kaiyang-Chen do you have time to review this change? |
Hi Cade. The original problem I see is that when enable prefix caching and use swap for the preemption, the
sure I can add smaller tests for the prefix_caching_block and native blocks |
|
lmk when it's ready for review @zachzzc |
Yes it is ready for review @cadedaniel |
| assert allocator_dst.get_num_blocks_touched( | ||
| src_blocks, num_lookahead_slots=1) == num_blocks | ||
| assert allocator_dst.get_num_blocks_touched( | ||
| src_blocks, num_lookahead_slots=block_size - 1) == num_blocks |
There was a problem hiding this comment.
Can we make num_lookahead_slots a test parameter to follow the convention here?
There was a problem hiding this comment.
The three lookahead situations are block_size dependent here, I think it is more straightforward to leave it this way.
There was a problem hiding this comment.
LGTM, can merge once @Kaiyang-Chen 's comments are addressed
github-actions
bot
added
the
ready
|
@simon-mo can we get a force merge here |
|
🫡 |
|
thanks for the contribution @zachzzc ! |
Signed-off-by: Alvant <[email protected]>
FILL IN THE PR DESCRIPTION HERE
FIX #xxxx (link existing issues this PR will resolve)
In the existing logics, the number of touched blocks is not calculated correctly.
If the prefix caching is enabled, and the blocks are swapped out, the left-over the caches will be in the evictor group if there are no references. In the meantime, the cache blocks will be evicted, for example, 2 out of 10 cached blocks. Let's say the remaining 8 blocks in the evictor are the last available blocks in the allocator, and the sequence wants to swapped in, the free blocks will count as 8. The touched blocks is calculated as 2, which is incorrect because the cache is not shared by any other blocks. In this case the correct touched blocks should be 10, and can't swap in.
In the regular case, if the lookahead slots are more than the free slots, it first need to occupy the block containing the free slots, then plus the remaining required blocks.
BEFORE SUBMITTING, PLEASE READ THE CHECKLIST BELOW AND FILL IN THE DESCRIPTION ABOVE
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,LLMclass, 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:
format.shto format your code.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-requiredand 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:
action-requiredlabel on the PR if there are changes required. The contributor should address the comments and ping the reviewer to re-review the PR.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!