Add the ability to specify the 'user' parameter by OpenAI dynamically when calling invoke on a chain

[niv-hertz]

Checked

• I searched existing ideas and did not find a similar one
• I added a very descriptive title
• I've clearly described the feature request and motivation for it

Feature request

Allow passing dynamic parameters to the chat completions call to the underlying LLM, when calling invoke from the top-level chain. Specifically, the user parameter to OpenAI/AzureOpenAI: https://platform.openai.com/docs/guides/safety-best-practices/end-user-ids

Motivation

The user parameter by OpenAI\AzureOpenAI allows specifying a user ID with the chat completions call, to be able to track users dynamically for the safety and security features. Currently (Unless I missed something), the only way to do that is using the kwargs when initializing the OpenAI/AzureOpenAI client. That means that in order to pass this parameter, we have to initialize a client for each of the users, which means initializing a chain for each user separately. This is quite cumbersome and requires storing a map of users to chains. It will be far easier to pass this parameter in the invoke call, and maintaining one chain.

Proposal (If applicable)

No response