Skip to content

Releases

v1.9.4

Choose a tag to compare

@comdaze comdaze released this 11 May 14:17

Generative BI Using RAG - Release Notes v1.9.4

中文

新功能

增强的 LLM 管理与 Bedrock Converse API 支持

  • 在 LLM 管理界面中添加了对 Amazon Bedrock Converse API 的支持
  • 在 utils/llm.py 中实现了 invoke_bedrock_converse 函数以处理 Converse API 调用
  • 通过 Bedrock Converse API 创建了统一的模型连接处理
  • 如果 Converse API 失败,添加了自动回退到传统 invoke_model API 的功能

用户自定义 AWS 凭证 (AK/SK) 支持

  • 添加了用户在创建或更新 Bedrock 模型时提供自己的 AWS 凭证的功能
  • 实现了将凭证以 JSON 格式安全存储在模型的 input_format 字段中
  • 增强了模型调用函数中的凭证处理
  • 在模型创建和更新表单中添加了凭证输入界面元素

改进

  • 增强了 get_query_intent 和 get_query_rewrite 函数中的错误处理
  • 改进了日志记录,以便更好地排查模型连接问题
  • 为模型调用失败添加了详细的错误消息
  • 通过更好的错误处理优化了模型响应提取

错误修复

  • 修复了查询处理函数中的错误处理
  • 解决了凭证验证和使用的问题
  • 修复了不同模型类型的响应格式处理
  • 改进了失败 API 调用的错误日志记录

技术细节

  • Converse API 实现为所有 Bedrock 模型提供了更一致的接口
  • 用户凭证以 JSON 格式安全存储在模型配置中
  • 错误处理现在包括详细的日志记录,便于故障排除
  • 模型调用现在支持自定义区域和凭证

English

New Features

Enhanced LLM Management with Bedrock Converse API Support

  • Added support for Amazon Bedrock Converse API in the LLM management interface
  • Implemented invoke_bedrock_converse function in utils/llm.py to handle Converse API calls
  • Created unified model connection handling through Bedrock Converse API
  • Added automatic fallback to traditional invoke_model API if Converse API fails

User-Defined AWS Credentials (AK/SK) Support

  • Added ability for users to provide their own AWS credentials when creating or updating Bedrock models
  • Implemented secure storage of credentials in the model's input_format field as JSON
  • Enhanced credential handling in model invocation functions
  • Added UI elements for credential input in model creation and update forms

Improvements

  • Enhanced error handling in get_query_intent and get_query_rewrite functions
  • Improved logging for better troubleshooting of model connection issues
  • Added detailed error messages for failed model invocations
  • Optimized model response extraction with better error handling

Bug Fixes

  • Fixed error handling in query processing functions
  • Resolved issues with credential validation and usage
  • Fixed response format handling for different model types
  • Improved error logging for failed API calls

Technical Details

  • The Converse API implementation provides a more consistent interface for all Bedrock models
  • User credentials are securely stored in JSON format in the model configuration
  • Error handling now includes detailed logging for easier troubleshooting
  • Model invocation now supports custom regions and credentials

v1.9.2

Choose a tag to compare

@comdaze comdaze released this 03 May 06:03

Generative BI using RAG on AWS - Release Notes

版本 1.9.2 (2025年5月3日)

新功能

  1. 增强的嵌入模型管理

    • 新的嵌入模型管理系统,支持多种嵌入平台(Bedrock、SageMaker、BR Client API)
    • 添加了embedding.py模块,用于全面的嵌入模型管理
    • 通过dynamo_embedding.py集成DynamoDB存储嵌入模型配置
    • 全局设置管理,支持设定默认的LLM模型、嵌入模型和数据配置文件
  2. 改进的实体识别

    • 增强实体检索,添加分数阈值(>0.8)以提高准确性
    • 实体信息现在保存到answer对象中,用于知识搜索
  3. SQL自动纠错

    • 为代理任务添加SQL自动纠错逻辑
    • 使用错误信息自动尝试修复失败的SQL查询
    • 为失败的SQL执行提供详细的错误日志

错误修复

  1. 向量存储嵌入改进

    • 增强嵌入生成中的错误处理
    • 添加对不同嵌入响应格式的支持
    • 实现嵌入失败的回退机制
    • 改进嵌入相关操作的日志记录
  2. 状态机工作流修复

    • 通过正确过渡实体检索,修复了知识搜索工作流
    • 改进了带有列表类型查询的代理任务处理
    • 当所有代理任务失败时提供更好的错误处理

架构改进

  1. 强大的错误处理

    • 在嵌入和查询处理管道中全面处理错误
    • 当服务返回意外格式时优雅回退
    • 提供详细日志以便故障排除
  2. 增强的查询处理

    • 改进知识搜索意图检测和处理
    • 实体检索和知识搜索之间更好的集成
  3. 模型集成

    • 灵活的模型配置系统,适用于LLM和嵌入模型
    • 支持不同模型提供商的各种输入/输出格式
  4. 全局设置管理

    • 支持设置默认LLM模型,方便统一管理
    • 支持设置默认嵌入模型,确保一致的向量表示
    • 支持设置默认数据配置文件,简化多配置文件环境
  5. 中英文双语支持

    • 增强了对中文和英文查询的处理能力
    • 改进了中文实体识别和语义理解
    • 优化了中英文混合环境下的系统响应

Generative BI using RAG on AWS - Release Notes

Version 1.9.2 (May-3-2025)

New Features

  1. Enhanced Embedding Model Management

    • New embedding model management system with support for multiple embedding platforms (Bedrock, SageMaker, BR Client API)
    • Added embedding.py module for comprehensive embedding model management
    • DynamoDB integration for storing embedding model configurations via dynamo_embedding.py
    • Global settings management for default LLM models, embedding models, and data profiles
  2. Improved Entity Recognition

    • Enhanced entity retrieval with score threshold (>0.8) for better accuracy
    • Entity information now saved to answer object for use in knowledge search
  3. SQL Auto-correction

    • Added SQL auto-correction logic for agent tasks
    • Automatically attempts to fix failed SQL queries using error information
    • Detailed error logging for failed SQL executions

Bug Fixes

  1. Vector Store Embedding Improvements

    • Enhanced error handling in embedding generation
    • Added support for different embedding response formats
    • Implemented fallback mechanisms for embedding failures
    • Better logging for embedding-related operations
  2. State Machine Workflow Fixes

    • Fixed knowledge search workflow by properly transitioning through entity retrieval
    • Improved handling of agent tasks with list-type queries
    • Better error handling when all agent tasks fail

Architecture Improvements

  1. Robust Error Handling

    • Comprehensive error handling throughout the embedding and query processing pipeline
    • Graceful fallbacks when services return unexpected formats
    • Detailed logging for troubleshooting
  2. Enhanced Query Processing

    • Improved knowledge search intent detection and processing
    • Better integration between entity retrieval and knowledge search
  3. Model Integration

    • Flexible model configuration system for both LLM and embedding models
    • Support for various input/output formats across different model providers
  4. Global Settings Management

    • Support for setting default LLM model for unified management
    • Support for setting default embedding model for consistent vector representations
    • Support for setting default data profile for simplified multi-profile environments
  5. Bilingual Support (English and Chinese)

    • Enhanced processing capabilities for both Chinese and English queries
    • Improved Chinese entity recognition and semantic understanding
    • Optimized system responses in mixed Chinese-English environments