Add support of the anthropic model #233
Conversation
git-hulk
force-pushed
the
feature/anthropic-llm
branch
from
259afe2 to
0c29587
Compare
git-hulk
force-pushed
the
feature/anthropic-llm
branch
from
0c29587 to
6ec94da
Compare
git-hulk
changed the title
|
Hi @dpasiukevich @mazas-google, I have implemented the first version of supporting the anthropic model, and have the basic test on my side. Could you please have a look when you get time? I would like to confirm if the direction is correct from the maintainer's perspective, and happy to modify the PR according to your feedback. |
|
/gemini review |
|
I will try do a proper review tomorrow. Thank you for the contribution! |
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request adds comprehensive support for Anthropic models via Vertex AI, introducing a new anthropic package to adapt the model.LLM interface. The implementation is well-structured, separating request building, response parsing, and the core client logic. It correctly handles both streaming and non-streaming generation and is accompanied by a solid set of tests, including good coverage for the streaming logic. I've identified one high-severity issue concerning the in-place modification of input data, which could lead to unexpected side effects. My detailed comment suggests a safer approach to ensure data immutability. Overall, this is a high-quality contribution.
git-hulk
force-pushed
the
feature/anthropic-llm
branch
from
dfa4215 to
802b98c
Compare
|
@dpasiukevich, could you please take a look at this PR when you get time? |
This commit integrates the changes from google#233, which adds comprehensive support for Anthropic's Claude models via the model.LLM interface. Key features: - Support for both streaming and non-streaming generation - Integration with Vertex AI, Anthropic API, and AWS Bedrock providers - Tool/function calling with proper schema conversion - Handles multiple content types (text, images, code execution) - Comprehensive test coverage Changes include: - New model/anthropic package with core implementation - Request/response builders for API translation - Provider selection (vertex_ai, anthropic, aws_bedrock) - Dependencies: anthropic-sdk-go v1.17.0 and tidwall utilities All code has been formatted with go fmt, passes go vet checks, and all tests pass successfully. Original PR: google#233 Author: git-hulk <[email protected]>
This document provides a detailed technical comparison of how the integrated PR google#233 code uses the Anthropic SDK versus how the Charm Fantasy library implements similar functionality. Key findings: - Both use the official anthropic-sdk-go correctly - Fantasy has more production features (caching, advanced streaming) - ADK-Go is simpler and well-suited for basic Vertex AI usage - Identifies potential improvements for future iterations This analysis helps understand the tradeoffs and implementation patterns for Anthropic SDK integration in Go.
2 participants
This PR closes #225.