Add support for parallel_tool_calls option when configuring Langchain::Assistant

[andreibondarev]

Is your feature request related to a problem? Please describe.
We'd like to enable better control of tool calling when using Langchain::Assistant. Some of the supported LLMs (Anthropic and OpenAI) let you modify whether parallel tool calls ("multiple tool calls") can be made or not. In some use-cases the Assistant must call tools sequentially hence we should be able to toggle that option on the Assistant instance.

Describe the solution you'd like
Similar to tool_choice enable the developer to toggle:

assistant = Langchain::Assistant.new(parallel_tool_calls: true/false, ...)
assistant.parallel_tool_calls = true/false

Tasks

• Langchain::Assistant::LLM::Adapters::Anthropic support
• Langchain::Assistant::LLM::Adapters::OpenAI support
• Langchain::Assistant::LLM::Adapters::GoogleGemini support (not currently supported)
• Langchain::Assistant::LLM::Adapters::MistralAI support (not currently supported)
• Langchain::Assistant::LLM::Adapters::Ollama support (not currently supported)

[sergiobayona]

I seems Google Gemini does support parallel function calling, see:

https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/function-calling#supported_models
https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/function-calling#parallel-samples

[andreibondarev]

I seems Google Gemini does support parallel function calling, see:

https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/function-calling#supported_models https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/function-calling#parallel-samples

It does, but there's no way to configure whether functions can be called in parallel or not.

[ms-ati]

If we pass parallel_tool_calls = false, could we stop the debug output each time telling us that the adapter doesn't support parallel tool calls?

[andreibondarev]

If we pass parallel_tool_calls = false, could we stop the debug output each time telling us that the adapter doesn't support parallel tool calls?

Have you tried changing the logger level? Something like Langchain.logger.level = Logger::ERROR should work.