server : avoid common_batch #16
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
ggml-ci
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the You can disable this status message by setting the 🪧 TipsChatThere are 3 ways to chat with CodeRabbit:
Note: Be mindful of the bot's finite context window. It's strongly recommended to break down tasks such as reading entire modules into smaller chunks. For a focused discussion, use review comments to chat about specific files and their changes, instead of using the PR comments. CodeRabbit Commands (Invoked using PR comments)
Other keywords and placeholders
CodeRabbit Configuration File (
|
ngxson
reviewed
Mar 20, 2025
examples/server/server.cpp
Outdated
| slot.batch_spec.clear(); | ||
| slot.batch_spec.add_text(id, slot.n_past, slot.id, true); | ||
| //slot.batch_spec.clear(); | ||
| //slot.batch_spec.add_text(id, slot.n_past, slot.id, true); |
There was a problem hiding this comment.
I think these comments can be safely removed since we no longer have notion of common_batch in the code base, these calls are specific to common_batch
ggerganov
force-pushed
the
xsn/private_batch_api_pooling_none
branch
from
5cebe54 to
b8b1732
Compare
|
Re. pooling == none, yes I completely agree that we can temporary disable it and reimplement in the future. For now it's only used by TTS so I guess it won't affect much. Merging this now, thanks! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
ref ggml-org#11875
Avoid usage of
common_batchinllama-server.The only use case that is not covered with this change is embeddings with pooling type ==
none. In this case, we have to output an embedding vector for each input token.With the current API, it is a little bit difficult to implement this, without using a helper struct such as
common_batch. Moreover, I think that onmasterthere is a bug where the tokens of a given slot could end up separated in different batches which would result in multiple partial results for a single task being emitted. This is probably very rare (requires parallel users and fragmented KV cache), so that's why we haven't experienced it yet.I am wondering if we should disable the option to have embeddings without pooling for now and figure out how to properly support this in the future.
@ngxson What do you think?