Add support for Bedrock Converse API (Anthropic Messages API, Claude 3.5 Sonnet)

[austintlee]

Description

Adds support for latest Message API

Related Issues

Resolves #2826

Check List

• [ x] New functionality includes testing.
• New functionality has been documented.
• API changes companion pull request created.
• [ x] Commits are signed per the DCO using --signoff.
• Public documentation issue/PR created.

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.
For more information on following Developer Certificate of Origin and signing off your commits, please check here.

[austintlee]

I need a jpeg file and a pdf file for testing image and document contents and they are failing the utf-8 content checks.

I see that in the core, the ingest plugin has this:

 forbiddenPatterns {
   exclude '**/*.doc'
   exclude '**/*.docx'
   exclude '**/*.pdf'
   exclude '**/*.epub'
   exclude '**/*.vsdx'
 }

I can add that to plugin/build.gradle.

[Zhangxunmt]

@austintlee just a few comments left but LGTM overall.
Since this new API support will be used by the requester in #2826, do you mind adding a tutorial for using this API like this one https://github.com/opensearch-project/ml-commons/blob/main/docs/tutorials/conversational_search/conversational_search_with_Cohere_Command.md, and update the limitations specified in this tutorial? I am 100% sure that the requester will follow the tutorial to build solutions.

[Zhangxunmt]

a leftover? remove this line?

[austintlee]

It's just for debugging. This is only in a test.

[Zhangxunmt]

Instead of keep adding new "else if" blocks here, why not just define an annotation for different methods and use reflection to process different models servers at runtime? In that way I think we could reduce the code complexity here.

[austintlee]

Given the time constraint, I would do the refactoring at a later time. I have been thinking about a good way to avoid this style of handling different cases of LLM vendors and APIs, but I was waiting for some general patterns to emerge which I think is this Message API, but again there are still small differences between LLM providers.

[austintlee]

@austintlee just a few comments left but LGTM overall. Since this new API support will be used by the requester in #2826, do you mind adding a tutorial for using this API like this one https://github.com/opensearch-project/ml-commons/blob/main/docs/tutorials/conversational_search/conversational_search_with_Cohere_Command.md, and update the limitations specified in this tutorial? I am 100% sure that the requester will follow the tutorial to build solutions.

Yes, I'll produce some documentation, but I don't have time this week. You can look at the IT tests to get a sense of the syntax for now if that helps with reviewing and making sense of the code.

[ylwu-amzn]

bldr is hard to understand , use more meaningful variable name?

[austintlee]

Sure.

[austintlee]

  {"error":{"root_cause":[{"type":"status_exception","reason":"Error from remote service: {\"message\":\"The security token included in the request is invalid.\"}"}],"type":"status_exception","reason":"Error from remote service: {\"message\":\"The security token included in the request is invalid.\"}"},"status":403}

[austintlee]

Is this an issue with GH setup?

[mingshl]

Do we need license for the picture or pdf in this PR? @kolchfa-aws

[mingshl]

@austintlee is this image generated from openai image generation?

[kolchfa-aws]

@austintlee Could you tell me the original source of this image please?

[austintlee]

FYI, that image is from OpenAI's API documentation - https://platform.openai.com/docs/api-reference/chat/create. It's in the example they give for including images in chat completion.

[austintlee]

The actual source is Wikipedia.

[Zhangxunmt]

@austintlee any objection to merge the PR if the issue is only in the test itself? Let’s use a separate PR to track the IT issue.

[austintlee]

RestBedRockInferenceIT > test_bedrock_multimodal_model FAILED
    org.opensearch.client.ResponseException: method [POST], host [http://[::1]:38749], URI [/_plugins/_ml/models/null/_deploy], status line [HTTP/1.1 404 Not Found]
    {"error":{"root_cause":[{"type":"status_exception","reason":"Failed to find model"}],"type":"status_exception","reason":"Failed to find model"},"status":404}

This is not my test. Is this a known issue? Some (internal) model got removed? Am I the only one seeing this issue?

[austintlee]

Also, all these time-outs we were seeing - I don't see this happening on Windows or Mac. Should I try to repro it on linux? It's really frustrating that I can't repro it locally.

[Zhangxunmt]

Also, all these time-outs we were seeing - I don't see this happening on Windows or Mac. Should I try to repro it on linux? It's really frustrating that I can't repro it locally.

That test flaky is because of the congestion in the ITs running environment. We should remove some outdated or duplicate ITs to unchoke. That multi modal IT should update to use auto deploy so it won’t fail due to model not found because it will be auto deployed in that case.

[austintlee]

Can we merge this?

[opensearch-trigger-bot]

The backport to 2.x failed:

The process '/usr/bin/git' failed with exit code 1

To backport manually, run these commands in your terminal:

# Fetch latest updates from GitHub
git fetch
# Create a new working tree
git worktree add .worktrees/backport-2.x 2.x
# Navigate to the new working tree
cd .worktrees/backport-2.x
# Create a new branch
git switch --create backport/backport-2851-to-2.x
# Cherry-pick the merged commit of this pull request and resolve the conflicts
git cherry-pick -x --mainline 1 17e81ae618bebc72b8d3cb76d57f1556b1d8c8e1
# Push it to GitHub
git push --set-upstream origin backport/backport-2851-to-2.x
# Go back to the original working tree
cd ../..
# Delete the working tree
git worktree remove .worktrees/backport-2.x

Then, create a pull request where the base branch is 2.x and the compare/head branch is backport/backport-2851-to-2.x.

[opensearch-trigger-bot]

The backport to 2.17 failed:

The process '/usr/bin/git' failed with exit code 1

To backport manually, run these commands in your terminal:

# Fetch latest updates from GitHub
git fetch
# Create a new working tree
git worktree add .worktrees/backport-2.17 2.17
# Navigate to the new working tree
cd .worktrees/backport-2.17
# Create a new branch
git switch --create backport/backport-2851-to-2.17
# Cherry-pick the merged commit of this pull request and resolve the conflicts
git cherry-pick -x --mainline 1 17e81ae618bebc72b8d3cb76d57f1556b1d8c8e1
# Push it to GitHub
git push --set-upstream origin backport/backport-2851-to-2.17
# Go back to the original working tree
cd ../..
# Delete the working tree
git worktree remove .worktrees/backport-2.17

Then, create a pull request where the base branch is 2.17 and the compare/head branch is backport/backport-2851-to-2.17.

[austintlee]

I opened #2914 to bring back the rest of the tests.

[austintlee]

Doc - opensearch-project/documentation-website#8195. I don't have a PR, yet. Do I need to prepare a PR or is it someone from the docs team?

[kolchfa-aws]

@austintlee Our workflow is that the feature developer puts up a documentation PR in the doc repo and then the doc team reviews. Please feel free to ask any questions. Thank you!