-
-
Notifications
You must be signed in to change notification settings - Fork 4.5k
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.
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
[Kernel] Correctly invoke prefill & decode kernels for cross-attention (towards eventual encoder/decoder model support) #4888
Conversation
@afeldman-nm Sorry for the delay. Let me take a look at the PR this afternoon. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@afeldman-nm Thanks for the PR.
Overall, I'm OK with this change. The code in xformers.py
is a bit difficult to follow, but it's understandable given the inherent complexity.
The only thing I'd like to ask is to avoid using \
in the code as guided in https://google.github.io/styleguide/pyguide.html#32-line-length Please use parentheses if the line overflows. Also, please considering adding ,
to the last argument of the methods (particularly in the forward
method of the attention backends) if that reduces the line changes.
nit: This will reduce the number of line changes and make the code look better. Co-authored-by: Woosuk Kwon <[email protected]>
Co-authored-by: Woosuk Kwon <[email protected]>
@WoosukKwon Thanks, I have addressed your feedback (both the specific points of feedback as well as the more general styleguide suggestsions.) I agree it is hard to make xformers comprehensible given the complexity, perhaps this can wait for a future PR. Please let me know if you have any other changes to recommend, or if the fixes look satisfactory. |
Thanks @sroy745 @maxdebayser @njhill @WoosukKwon @robertgshaw2-neuralmagic for all of your excellent help in landing this PR! |
…n (towards eventual encoder/decoder model support) (vllm-project#4888) Co-authored-by: Woosuk Kwon <[email protected]>
…n (towards eventual encoder/decoder model support) (vllm-project#4888) Co-authored-by: Woosuk Kwon <[email protected]>
…n (towards eventual encoder/decoder model support) (vllm-project#4888) Co-authored-by: Woosuk Kwon <[email protected]> Signed-off-by: Alvant <[email protected]>
This PR is a step towards encoder/decoder model support. This PR modifies the xFormers backend* such that (1) the attention impl can implement cross-attention, and (2) the attention metadata data structure can represent the necessary metadata for invoking cross-attention.
* FlashAttention backend support for encoder/decoder models is left as future work
A quick overview of the plan for supporting encoder/decoder models in vLLM:
Prefill phase: (1) Non-autoregressive encoder inference yields encoder hidden states in a single pass; no KV caching occurs. (2) decoder prefill yields first-token-prediction & cached KVs. Within the decoder, cross-attention layers cache the KVs derived from encoder hidden states:
Key_{cross-attn, layer-n} = W_{K, cross-attn, layer-n} x (Encoder hidden states)
Value_{cross-attn, layer-n} = W_{V, cross-attn, layer-n} x (Encoder hidden states)
Note that all cross-attention layers consume the same encoder hidden states; however each cross-attention layers' keys and values differ because each layer has unique W_{K, cross-attn, layer-n} and W_{V, cross-attn, layer-n}. Therefore, the cross-attention KV cache must store KVs for each decoder layer, even though these KVs are all derived from a single set of encoder hidden states.
Note that self-attention layer behavior is unchanged compared to what it would be in a decoder-only model (cache KVs computed from the previous decoder layer outputs.)
Decode phase: during each iteration of the autoregressive decode process,
To implement the above encoder/decoder inference process, the following functionality will be added to vLLM over the course of multiple PRs:
Note 1: because this PR makes an incremental contribution (cross-attention KV-caching and memory management), this PR will not enable end-to-end encoder/decoder support (this will rely on later PRs.)
Note 2: the best effort is being made to ensure that encoder/decoder models are compatible with existing vLLM features. At this time, encoder/decoder models are unlikely to be compatible with the following vLLM features:
INCREMENTAL FIX TOWARDS #187
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,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:
format.sh
to 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-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:
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.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!