server: use chunked inputs #1985
Merged
Conversation
This file contains 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
danieldk
force-pushed
the
feature/server-chunks
branch
2 times, most recently
from
544ddf5 to
64bea24
Compare
5 tasks
danieldk
force-pushed
the
feature/server-chunks
branch
from
64bea24 to
94affb7
Compare
| from text_generation_server.pb import generate_pb2 | ||
|
|
||
|
|
||
| def concat_text_chunks(chunks: Iterable[generate_pb2.InputChunk]) -> str: |
There was a problem hiding this comment.
Is this method only to be future proof or is there a way today to have multiple text chunks?
There was a problem hiding this comment.
AFAIK we can currently only have multiple text chunks in a VLM models, so this was indeed only to future proof.
There was a problem hiding this comment.
Then mb we should take [0] and crash if len > 1 with an unreachable?
There was a problem hiding this comment.
Oh, we do need to iterate over chunks, because we are sending image chunks unconditionally during warmup, even for text-only models:
The current approach seems more robust? What do you think about logging a warning when len(texts) > 1?
There was a problem hiding this comment.
Updated to:
- Fail when there is more than one text chunk.
- Fail when there is no text chunk.
- Log at debug-level when there is a non-text chunk (only log because e.g. warmup sends an image chunk).
drbh
reviewed
Jun 3, 2024
danieldk
force-pushed
the
feature/server-chunks
branch
from
94affb7 to
77a6805
Compare
danieldk
force-pushed
the
feature/server-chunks
branch
from
77a6805 to
e1dd4ae
Compare
OlivierDehaene
previously approved these changes
Jun 6, 2024
danieldk
force-pushed
the
feature/server-chunks
branch
from
e1dd4ae to
3181443
Compare
The router will now send the input as chunks besides as a single string. This change modifies the server to process chunked input rather than strings. This also allows us to remove the image extraction code from the server.
danieldk
force-pushed
the
feature/server-chunks
branch
from
3181443 to
633e2bd
Compare
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.
What does this PR do?
The router will now send the input as chunks besides as a single string. This change modifies the server to process chunked input rather than strings. This also allows us to remove the image extraction code from the server.
Draft note: mostly checking whether all models pass, also needs to be rebased on #1981 once that PR is merged.
Before submitting
Pull Request section?
to it if that's the case.
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.