Future work suggestions

[maciejmajek]

Hello,
Here are some suggestions that I've gathered while using the vendors-gem:

Claude/Anthropic/Bedrock

• The user is permitted to specify a temperature value greater than 0 and less than 1. If the model is prompted with a flawed configuration, a pop-up appears. Considering that temperature is a parameter strictly bound between 0 and 1, perhaps implementing a slider would be effective? Same with Top_P

• Setting stop sequence results in Error: Malformed input request: #: extraneous key [stop_sequence] is not permitted, please reformat your input and try again.

• anthropic Model Ids available at bedrock (🟢 - works, 🔴 does not work)

  • 🟢 anthropic.claude-instant-v1
  • 🟢 anthropic.claude-v2
  • 🟢 anthropic.claude-v2:1
  • 🔴 anthropic.claude-3-haiku-20240307-v1:0 Error: Could not resolve the foundation model from the provided model identifier.
  • 🔴 anthropic.claude-3-sonnet-20240229-v1:0 Error: Could not resolve the foundation model from the provided model identifier.
  • **The prompt format changes between the v2 and 3 model. Consider changing the text box to a list of supported models (or two lists: first list for choosing anthropic/cohere/amazon/mistral, second list for choosing supported model) ** link (see #API request)

• As previously mentioned, different models hosted on AWS Bedrock may utilize different prompt formats. Therefore, naming the requester 'AWS SDK Requester component' might be misleading, especially when it's actually a Claude/Anthropic requester. Additionally, the Claude/Anthropic requester currently does not support all available Anthropic models.
  However, I believe that if the requester is enhanced with different configurations for various vendors and models, changing the name may not be necessary.

• Consider using list with allowed keyboard input for Region Name

• I find setting model Id in requester odd, especially when there is another tab called 'Model Configurations' which in case of ollama holds both model name and parameters. I believe that replacing requesters with 'Backend' might be a sensible option.

  • AWS Backend
    • Region Name
  • Ollama Backend
    • Endpoint ip:port/
    • Alternatively list of choices (/api/chat; api/generate)

Ollama

• Naming. BasicJSONRequesterComponent. Is this component reusable for other backends/vendors?

• What is "Content type" in BasicJSONRequesterComponent confiuration?
  https://github.com/RobotecAI/ai-vendors-gems/blob/0cafcfa0cc8056d9cbb44eeef45837613264a1b9/Gems/GenAIOllama/Code/Source/Communication/JSONHttp/BasicJSONRequesterComponent.cpp#L43-L47

• Consider changing the following into "An url with port pointing to an HTTP endpoint (format address:port/endpoint_url)"
  https://github.com/RobotecAI/ai-vendors-gems/blob/0cafcfa0cc8056d9cbb44eeef45837613264a1b9/Gems/GenAIOllama/Code/Source/Communication/JSONHttp/BasicJSONRequesterComponent.cpp#L42

• Model configuration for ollama models

  • Consider adding a link pointing to ollama library for users' convenience https://ollama.com/library
  • Consider adding similar to claude ui for setting basic parameters such as temperature, topp, topk

[jhanca-robotecai]

I transferred this issue from ai-vendor-gems repo due to the code transfer. All comments in this issue should be addressed in ai-core-gem repository.