[Hardware][Intel] fp8 kv cache support for CPU #5492
Conversation
cmake/cpu_extension.cmake
Outdated
There was a problem hiding this comment.
I think we want to compile for AVX512BF16 if we simply find it, not if we find it and force it
csrc/cpu/attention.cpp
Outdated
There was a problem hiding this comment.
Why aren't these asserts valid anymore?
csrc/cpu/attention.cpp
Outdated
There was a problem hiding this comment.
It isn't immediately clear that cpu_fp8 is treated as a type since the PyTorch types use PascalCase - maybe at least adding a _t such as cpu_fp8_t or CpuFP8 or FP8Cpu?
csrc/cpu/attention.cpp
Outdated
There was a problem hiding this comment.
Why do you need this IS_FP8_KV_CACHE boolean in paged_attention_v1 but not in paged_attention_v2? It seems like it could be deduced by checking the cache_t within the kernel and this would bode better for future cache types being used.
csrc/cpu/cpu_types.hpp
Outdated
There was a problem hiding this comment.
Calling out again that I think the name of this type doesn't seem like a type.
csrc/cpu/fp8_utils.h
Outdated
There was a problem hiding this comment.
For all these function signatures that have just e5m2 to signify fp8e5m2, I think it would be more clear to include the fp8
csrc/cpu/fp8_utils.h
Outdated
There was a problem hiding this comment.
Any additional function header or inline comments would be appreciated!
csrc/cpu/fp8_utils.h
Outdated
There was a problem hiding this comment.
What is the difference between noinf and noINF?
csrc/cpu/fp8_utils.h
Outdated
There was a problem hiding this comment.
It seems you aren't getting any reuse out of these functions, why not inline the body?
There was a problem hiding this comment.
This is only implemented for CPU usage of the Torch SPDA backend, right? Would fp8 kv cache work if we used CUDA with the SPDA backend, for instance?
csrc/cpu/attention.cpp
Outdated
There was a problem hiding this comment.
It'd be better to align with the general interface. We now support fp8, fp8_e4m3 and fp8_e5m2. Note that fp8 defaults to fp8_e4m3. Please cover all cases and throw an error for unsupported cases.
jikunshang
force-pushed
the
cpu_fp8_kv_cache
branch
from
b8b94af to
174c369
Compare
|
@jikunshang do you plan to continue this work? |
@bigPYJ1151 Please help update status, thanks! |
|
@hmellor I think this PR can be closed as it has been integrated to the intel-extension-for-pytorch 2.6. The feature will be enabled after the torch 2.6 update finished. |
|
@bigPYJ1151 thanks! Let me close this one. |
|
Great! In that case, could you remove the TODO from the docs regarding this feature? |
|
@hmellor sure, will update the doc too. |
6 participants
FILL IN THE PR DESCRIPTION HERE
Add FP8 kv cache for Intel CPU.
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!