docs: discussion recipe models relation

[axel7083]

See docs/proposals/recipe-models-relation.md

Related to #1111

[jeffmaury]

My 2 cents:

• the cuda and other variants shoud be handled by the inferenceprovider based on the environment
• inference providers should be registered according to the backend they support
• I don't see the use case for llama.cpp and ollama as they are the same
• I'm +1 for backend property

[jeffmaury]

See comment

[axel7083]

My 2 cents:

• the cuda and other variants shoud be handled by the inferenceprovider based on the environment

Okey with this proposition

• I don't see the use case for llama.cpp and ollama as they are the same

When I say llama.cpp I am thinking about our own image build by the recipe team. Because they probably do not need the same arguments, environment variables etc. They should be consider as two separate provider.

• I'm +1 for backend property

Good, then #1186

[axel7083]

Closing as resolved