Feat(amazon bedrock): Add support for cohere embed models #6190
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
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…nge (vercel#5531) Co-authored-by: Carl Brugger <[email protected]>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…#5541) Co-authored-by: Bram Meerten <[email protected]>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
) Co-authored-by: Sam Denty <[email protected]> Co-authored-by: Sam Denty <[email protected]>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
## Background This resolves an issue a user ran into: vercel#6041 ## Summary Aligned openai provider with openai-compatible
## Background This pull request adds the ability for clients to resume an ongoing chat generation stream after a network disconnect. ## Summary This pull request adds support for the `useChat` hook to resume an ongoing chat generation stream by exposing `experimental_resume()` that can be called by any client, typically during the initial mount of the hook. The `experimental_resume` function makes a `GET` request to the api endpoint you've initialized the hook with (or `/api/chat` by default) and streams the contents of the stream if it is active or fails silently if it has ended. In order for `experimental_resume` to work as intended, it requires the usage of the [`resumable-stream`](https://www.npmjs.com/package/resumable-stream) package for stream creation and a redis instance for the package to manage the pub/sub mechanism. ## Verification An example has been added at `examples/next-openai/app/use-chat-resume` to test the feature. Follow the following steps to test it end-to-end: 1. Run the development server 2. Navigate to `http://localhost:3000/use-chat-resume` 3. Send a message that will have a longer generation duration, example "Write an essay about Michael Jordan" 4. Once the generation starts, click the chat id above to open the conversation in a new tab 5. Verify the stream gets resumed ## Tasks - [x] Tests have been added / updated (for bug fixes / features) - [x] Documentation has been added / updated (for bug fixes / features) - [x] A _patch_ changeset for relevant packages has been added - [x] Formatting issues have been fixed (run `pnpm prettier-fix` in the project root) --------- Co-authored-by: Lars Grammel <[email protected]>
# Releases ## [email protected] ### Patch Changes - Updated dependencies [6c59ae7] - @ai-sdk/[email protected] - @ai-sdk/[email protected] ## @ai-sdk/[email protected] ### Patch Changes - 6c59ae7: feat (ui/react): support resuming an ongoing stream - Updated dependencies [6c59ae7] - @ai-sdk/[email protected] ## @ai-sdk/[email protected] ### Patch Changes - Updated dependencies [6c59ae7] - @ai-sdk/[email protected] ## @ai-sdk/[email protected] ### Patch Changes - Updated dependencies [6c59ae7] - @ai-sdk/[email protected] ## @ai-sdk/[email protected] ### Patch Changes - 6c59ae7: feat (ui/react): support resuming an ongoing stream ## @ai-sdk/[email protected] ### Patch Changes - [email protected] ## @ai-sdk/[email protected] ### Patch Changes - Updated dependencies [6c59ae7] - @ai-sdk/[email protected] Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
## Background This pull request fixes a docs formatting issue due to a missing comma in useChat reference docs. ## Summary Fixed formatting in useChat reference docs by adding a comma. ## Tasks - [x] Documentation has been added / updated (for bug fixes / features) - [x] Formatting issues have been fixed (run `pnpm prettier-fix` in the project root) ## Related Issues vercel#6052
## Background The Dify provider allows developers to easily integrate Dify's application workflow with their applications using the Vercel AI SDK. ## Summary This PR adds the Dify provider documentation to the AI SDK, enabling users to understand how to implement and use the Dify integration. ## Tasks - [x] Documentation has been added for the Dify provider - [x] Installation instructions are included - [x] Usage examples for both cloud and self-hosted instances are provided ## Related Issues [Support Dify Provider](vercel#4084) --------- Co-authored-by: Lars Grammel <[email protected]>
<!-- Welcome to contributing to AI SDK! We're excited to see your changes. We suggest you read the following contributing guide we've created before submitting: https://github.com/vercel/ai/blob/main/CONTRIBUTING.md --> ## Background <!-- Why was this change necessary? --> ## Summary <!-- What did you change? --> ## Verification <!-- For features & bugfixes. Please explain how you *manually* verified that the change works end-to-end as expected (independent of automated tests). Remove the section if it's not needed (e.g. for docs). --> ## Tasks <!-- This task list is intended to help you keep track of what you need to do. Feel free to add tasks and remove unnecessary tasks as needed. Please check if the PR fulfills the following requirements: --> - [ ] Tests have been added / updated (for bug fixes / features) - [ ] Documentation has been added / updated (for bug fixes / features) - [ ] A _patch_ changeset for relevant packages has been added (for bug fixes / features - run `pnpm changeset` in the project root) - [ ] Formatting issues have been fixed (run `pnpm prettier-fix` in the project root) ## Future Work <!-- Feel free to mention things not covered by this PR that can be done in future PRs. Remove the section if it's not needed. --> ## Related Issues <!-- List related issues here, e.g. "Fixes vercel#1234". Remove the section if it's not needed. -->
## Background The safetyRating was an empty object for some reason ## Summary Fixes vercel#6109
…el#6133) ## Background Before in issues like vercel#5499 & vercel#5833, if users ran into an issue where the messages array failed to parse correctly they'd run into a difficult to debug error `Invalid prompt: messages must be an array of CoreMessage or UIMessage` ## Summary This PR extends these error messages to include more details on what exactly went wrong, which is useful information for debugging when stuff breaks. On security: I think this is an okay change since the data sent is not private
This adds gemini-2.5-pro-preview-05-06 to AI SDK
- Added support for Cohere embedding models with customizable settings including input type, truncation, and embedding types. - Implemented separate embedding methods for Cohere and Titan models within the BedrockEmbeddingModel class. - Updated response handling to accommodate both simple and complex embedding formats for Cohere. - Revised the BedrockEmbeddingSettings interface to include Cohere-specific settings.
- Introduced unit tests for the `doEmbed` method, covering both Cohere and Titan models. - Enhanced test cases to validate handling of single and multiple input values, image inputs, and various embedding types. - Updated server response mocks to reflect new embedding scenarios.
…lution - Added compiler options for module resolution and module type to support Node16 in the TypeScript configuration file, specific to bedrock sdk requirements.
- Added new export types for BedrockEmbeddingModelId, BedrockEmbeddingSettings, and CohereEmbeddingSettings in the index file
lgrammel
reviewed
May 7, 2025
|
Please add a changeset. |
lgrammel
reviewed
May 7, 2025
| /** | ||
| * Settings specific to Cohere embedding models. | ||
| */ | ||
| cohere?: CohereEmbeddingSettings; |
There was a problem hiding this comment.
Not a fan of this solution and the overall way of how the functionality is integrated. Need to think more about how to do this.
There was a problem hiding this comment.
Could you expand a bit more on what you would like to see? I can add more type safety for sure. But are you thinking more just rethinking the overall integration as a whole? I could split cohere and titan into their own class and then just set the general bedrockembeddingmodel to be a thin wrapper for the subclasses.
lgrammel
reviewed
May 7, 2025
| /** | ||
| * Specifies the types of embeddings you want to have returned. | ||
| */ | ||
| embedding_types?: Array<'float' | 'int8' | 'uint8' | 'binary' | 'ubinary'>; |
There was a problem hiding this comment.
We only support float atm.
lgrammel
reviewed
May 7, 2025
|
Ideally this would be on the |
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.
Background
The current Amazon Bedrock provider implementation doesn't properly support Cohere embedding models, which use a different request format compared to Titan models. This causes issues when attempting to use Cohere models for embeddings through the Amazon Bedrock provider.
Summary
bedrock-embedding-model.tsto handle Cohere-specific request and response formatsBedrockEmbeddingSettingstype in the package's index file to fix TypeScript errors when using the updated package in third-party applicationsVerification
I tested all changes using the automated test suite and verified functionality with each Bedrock embedding model (including Cohere models) in a Next.js 15 app router application using the updated dist outputs. All tests passed successfully, confirming that both Titan and Cohere embedding models now work as expected.
Tasks
pnpm prettier-fixin the project root)Related Issues
Fixes #5055