GML-2008 GML-2009 Azure Library Upgrades

[chengbiao-jin]

PR Type

Enhancement

---

Description

• Migrate Azure imports to langchain_openai.

• Add Bedrock botocore client configuration.

• Support token_limit for Azure/Google LLMs.

• Default model/region for Bedrock embeddings.

---

Diagram Walkthrough

flowchart LR
  A["Azure OpenAI services"]:::n --> B["Use langchain_openai"]:::n
  C["Bedrock services"]:::n --> D["botocore client config"]:::n
  E["LLMs: Azure & Google"]:::n --> F["Optional token_limit"]:::n
  G["Embeddings: Azure/AWS"]:::n --> H["Model/dimensions & region defaults"]:::n

  classDef n fill:#eef,stroke:#55f,stroke-width:1px

Loading

File Walkthrough

	Relevant files
Enhancement	embedding_services.py Azure embeddings migration and Bedrock client tuning          common/embeddings/embedding_services.py Swap to langchain_openai Azure embeddings. Pass model and dimensions explicitly. Add botocore client config for Bedrock. Default Bedrock model and configurable region. +13/-5    azure_openai_service.py AzureChatOpenAI migration and token limit support                common/llm_services/azure_openai_service.py Migrate to langchain_openai AzureChatOpenAI. Add optional max_tokens from token_limit. +2/-1      google_genai_service.py Google GenAI token limit configuration                                      common/llm_services/google_genai_service.py Use token_limit for max_tokens. Keep other model settings unchanged. +1/-1
Configuration changes	aws_bedrock_service.py Bedrock LLM boto3/botocore client configuration                    common/llm_services/aws_bedrock_service.py Import botocore and build client config. Configure retries, timeouts, and pool size. Pass config and region to Bedrock client. +10/-1

---

[tg-pr-agent]

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

⏱️ Estimated effort to review: 3 🔵🔵🔵⚪⚪

🧪 No relevant tests

🔒 No security concerns identified

⚡ Recommended focus areas for review Possible Issue AzureOpenAIEmbeddings from langchain_openai likely expects the azure_deployment argument (not deployment). Passing deployment may raise a constructor error; also verify whether model is needed when using Azure deployments. from langchain_openai import AzureOpenAIEmbeddings self.embeddings = AzureOpenAIEmbeddings(model=self.model_name, dimensions=self.dimensions, deployment=config["azure_deployment"]) Possible Issue AzureChatOpenAI from langchain_openai typically uses model (not model_name). Using model_name= can result in an unexpected keyword argument error. Also confirm that max_tokens is supported in the installed version. self.llm = AzureChatOpenAI( azure_deployment=config["azure_deployment"], openai_api_version=config["openai_api_version"], model_name=config["llm_model"], max_tokens=config.get("token_limit"), temperature=config["model_kwargs"]["temperature"], ) Parameter Name Mismatch ChatGoogleGenerativeAI commonly uses max_output_tokens rather than max_tokens. If max_tokens is unsupported, the limit may be ignored or raise an error. Verify against the installed langchain-google-genai version. self.llm = ChatGoogleGenerativeAI( temperature=config["model_kwargs"]["temperature"], model=model_name, max_tokens=config.get("token_limit"), timeout=None, max_retries=2, )