router: send the input as chunks to the backend #1981
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/chunked-input
branch
from
78bb569 to
99d4c9e
Compare
Narsil
reviewed
May 31, 2024
Yeah #1985 will do that. |
danieldk
force-pushed
the
feature/chunked-input
branch
from
99d4c9e to
6c5598e
Compare
danieldk
commented
May 31, 2024
router/client/src/lib.rs
Outdated
Comment on lines
74
to
76
| // We don't create empty chunks, probably better to be robust | ||
| // than unreachable!(). | ||
| None => {} |
There was a problem hiding this comment.
Just wanted to make sure you see this. Since we are creating the chunks in the router, this case should never occur. But we could also have unreachable!() or bubble up an error. I don't have a strong opinion.
There was a problem hiding this comment.
Either is fine.
Seems to be an artifact of the protobuf definition.
unreachablemakes thing hard crash so easier to debug, but less robust- do nothing, harder to debug should anything go wrong (if even visible) but more robust.
I don't a strong opinion to be honest. Insinctively I'd make it a unreachable or equivalent, at the very least in the silent path I would add a warn log.
danieldk
force-pushed
the
feature/chunked-input
branch
from
6c5598e to
fc52ba6
Compare
Before this change, the generation input was sent to the backend as a single string, encoding images as Base64 and packing them in Markdown-style links. This change adds a new chunked input representation that separates text chunks from images chunks. Image chunks contain binary data (for smaller message sizes) and the image's MIME type. The stringly-typed inputs are still sent to support backends that do not support chunked inputs yet.
danieldk
force-pushed
the
feature/chunked-input
branch
from
fc52ba6 to
f92411a
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?
Inputs are currently sent to the backend as a single string, encoding images as Base64 and packing them in Markdown-style links. We would like to switch to chunked inputs, with different chunk types corresponding to different modalities. This reduces the amount of input-parsing that needs to be done and increases safety by properly typing inputs. This change can be broken up in various steps:
This change implements the first step, adding a new chunked input representation that separates text chunks from images chunks. Image chunks contain binary data (for smaller message sizes) and the image's MIME type.
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.