Support o-models

[marcorosa]

OpenAI models o1 and o3-mini are available in AI Core and we can integrate them.

TBD

The following points require some attention in development and tests

• generative-ai-hub-sdk: these models will be officially supported starting from v4.7.0, but do we need such version? In the end we delegate the completions to our adapter in llm.py
  • v4.7.0 may not be sufficient -> ref to my issue open
• Remember these models do not support temperature nor many other parameters of regular gpt models, and have the max_tokens param renamed to max_completion_tokens
• llm.py: as these models fall under AICoreOpenAILLM we may need to adapt the generate_completions_for_messages function
  • either replace these parameters depending on the model name
  • or extend the base class so to override generate_completions_for_messages and pass only supported models
  • an upgrade of the generative-ai-hub-sdk library should be sufficient to have a working ProxyOpenAI (i.e., gen_ai_hub.proxy.native.openai.OpenAI) class with all openai's models.
    • this class is defined and used for AICoreOpenAILLM which is also adopted for all openai-compatible models (i.e., ibm, opensource (mistral+llama), and mistral). They all need to be tested afterwards
    • LocalOpenAILLM extends AICoreOpenAILLM but override generate_completion_for_messages. Still, validate no issues are raised

[marcorosa]

Done in #88, will be part of v0.5.0 release