feat: optimize kv cache load/offload. #306
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Kang-Meng
requested review from
RobbieLeung,
liutongxuan,
walsonyang and
yq33victor
Kang-Meng
force-pushed
the
feat_async_copy
branch
3 times, most recently
from
17797ce to
034f86e
Compare
RobbieLeung
reviewed
Nov 5, 2025
Kang-Meng
force-pushed
the
feat_async_copy
branch
2 times, most recently
from
7cd5bd4 to
d4446aa
Compare
liutongxuan
reviewed
Nov 6, 2025
liutongxuan
reviewed
Nov 6, 2025
Kang-Meng
force-pushed
the
feat_async_copy
branch
5 times, most recently
from
90457da to
b996af0
Compare
yq33victor
reviewed
Nov 9, 2025
| "", | ||
| "The address of the kv cache store metadata service."); | ||
|
|
||
| DEFINE_string(store_local_hostname, |
Collaborator
There was a problem hiding this comment.
what's the different between store_metadata_server and store_local_hostname.
| } | ||
| } | ||
| if (is_completed) { | ||
| close_future.wait(); |
Collaborator
There was a problem hiding this comment.
nit: If is_completed was set to false above, does that mean we no longer need to wait() on close_future here?
how brpc to handle stream_handler in this case
And by the way, how can we ensure that multiple batches are delivered in order or received in order?
| int Stream::synchronize() const { | ||
| #if defined(USE_NPU) | ||
| return aclrtSynchronizeStream(stream_.stream()); | ||
| return aclrtSynchronizeStreamWithTimeout(stream_.stream(), timeout_); |
Collaborator
There was a problem hiding this comment.
in which case we need timeout? and what happen if timeout.
| std::move(copy_out_blocks_async(input.input_params))); | ||
| { | ||
| std::lock_guard<std::mutex> lock(mutex_); | ||
| if (layer_wise_load_synchronizer_.count(input.input_params.batch_id) != |
Collaborator
There was a problem hiding this comment.
nit: can we dont use lock here ? just a suggestion
Kang-Meng
force-pushed
the
feat_async_copy
branch
5 times, most recently
from
629d719 to
1c0f9a2
Compare
walsonyang
reviewed
Nov 12, 2025
| std::vector<BlockTransferInfo> block_transfer_info; | ||
| proto_to_block_transfer_info(*req, block_transfer_info); | ||
|
|
||
| copy_threadpool_.schedule( |
Kang-Meng
force-pushed
the
feat_async_copy
branch
from
1c0f9a2 to
e5ba51f
Compare
walsonyang
reviewed
Nov 12, 2025
| proto_to_block_transfer_info(*req, block_transfer_info); | ||
|
|
||
| auto future = worker_->load_kv_blocks_from_store_async(dst_blocks); | ||
| BThreadPool::get_instance().schedule( |
Collaborator
There was a problem hiding this comment.
no need BThreadPool here, just bthread_start_background, bthread.join
Kang-Meng
force-pushed
the
feat_async_copy
branch
from
e5ba51f to
fb9d1c4
Compare
Kang-Meng
force-pushed
the
feat_async_copy
branch
from
fb9d1c4 to
8c05d99
Compare
Kang-Meng
force-pushed
the
feat_async_copy
branch
from
8c05d99 to
82627cd
Compare
Kang-Meng
force-pushed
the
feat_async_copy
branch
2 times, most recently
from
88b12ae to
b9cc794
Compare
walsonyang
previously approved these changes
Nov 14, 2025
Kang-Meng
force-pushed
the
feat_async_copy
branch
from
b9cc794 to
49d7280
Compare
Kang-Meng
force-pushed
the
feat_async_copy
branch
from
49d7280 to
881bfaa
Compare
yq33victor
previously approved these changes
Nov 17, 2025
Kang-Meng
force-pushed
the
feat_async_copy
branch
from
881bfaa to
86c9bc0
Compare
RobbieLeung
approved these changes
Nov 17, 2025
walsonyang
approved these changes
Nov 17, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.