Release v3.1.33

What's Changed

• anthropic: support caching
• improve vector stores, fix potentially not disposing httpresult
• gemini-2.0-flash-thinking-exp, gemini-2.0-flash-exp
• adding test case for custom chunking strategy
• Update README.md
• Update README.md
• fix
• Update project versions to v3.1.32
• minor changes
• adding self-cleaning test-cases
• fixing issues from PR review
• fixing issues from PR review
• fixing xml doc and pointing created_at to created property in base result class
• adding rest of the endpoints for VectorStoreFile
• fixing json convertor
• adding VectorStoreFile object and converter for chunking strategy
• adding vectore store to the baes API + changing json tag
• adding base for VectoreStore endpoint
• using Path.Join to file navigations rather Window's \

Updated Projects

LlmTornado.Demo.csproj -> 0.0.1
LlmTornado.csproj -> 3.1.33

Release v3.1.32

What's Changed

• add ollama streaming demo
• add ollama demo
• Update dotnet.yml
• Update dotnet.yml
• Update dotnet.yml
• fix
• exclude forks from status checking
• fix ci
• Pruned .gitignore
• Update .gitignore
• Update project versions to v3.1.31

Updated Projects

LlmTornado.Demo.csproj -> 0.0.1
LlmTornado.csproj -> 3.1.32

Release v3.1.31

What's Changed

• add release script
• fix
• fix
• setup ci
• llama-3.3-70b-versatile, llama-3.1-8b-instant, llama-guard-3-8b

Updated Projects

LlmTornado.csproj -> 3.1.31
LlmTornado.Demo.csproj -> 0.0.1

3.1.30

Features

• SSE implementation for OpenAI and Anthropic is reworked, fixing some rare issues with mid-response tool calls.

3.1.26

Breaking changes

• LlmTornado was refactored to drop enums.net dependency. Please install LlmTornado.Contrib to get back some FunctionCall helper methods.

Models

• OpenAI: gpt-4o-2024-11-20

Features

• Network-level safe APIs now expose the HttpContent of the outbound request and other details.
• Strict JSON mode is now supported in Google APIs.
• Library dependencies are now marked as ranges for increased compatibility.

3.0.18

Breaking changes

• Audio endpoint (transcription, translation) is undergoing an extensive rework to bring it up to the chat endpoint standard.
• Some of the old methods had the suffix Async removed.

Models

• OpenAI: gpt-4o-audio-preview
• Anthropic: claude-3-5-sonnet-20241022, claude-3-5-haiku-20241022
• Groq: whisper-large-v3-turbo, distil-whisper-large-v3-en, whisper-large-v3

Features

• Chat audio modality is now fully supported, including streaming, smart compression. All modes (audio in, text out; text in, audio out; audio in, audio out; mixed) are supported.
• Mapped missing bits from OpenAI, including new usage details, store and metadata fields.
• Extensive audio rework, see the new demos for more details.

3.1.5

Fixes

• Conversation.StreamResponse - fixed yielding one time more than necessary

Features

• OpenAI: gpt4-o-mini
• Groq: gemma2, mixtral7x8, llama3
• Cohere: embedding extensions (input type, truncation method)
• Anthropic: claude-3-5-sonnet

3.1.0

Breaking changes:

• ChatModel.Cohere.CommandRPlus -> ChatModel.Cohere.Claude3.CommandRPlus

Features

• Support Google Gemini models (text, [parallel] tools, streaming).
• Improve performance when dealing with rich responses (images).
• Support missing Cohere models: command, command-nightly, command-light, and command-light-nightly.

3.0.17

• fix parallel tool calls setting

3.0.16

• improved Anthropic streaming
• support disabling parallel tools calling