OpenAI Embeddings model used no matter the platform defined on the indexed

[welcoMattic]

Given this user configuration:

llm_chain:
    embedder:
        default:
            platform: 'llm_chain.platform.mistral'
            model:
                name: 'Embeddings'
                version: 'mistral-embed'

If the embeddings model name is "Embeddings" (as is the case for OpenAI, Mistral and Google), the model object injected in Embedder object will always be PhpLlm\LlmChain\Platform\Bridge\OpenAI\Embeddings due to https://github.com/php-llm/llm-chain-bundle/blob/main/src/DependencyInjection/LlmChainExtension.php#L462

I see 3 solutions to fix this issue:

• Force to (prefix|suffix) Embeddings model name with the platform name (e.g. MistralEmbeddings, OpenAIEmbeddings)
• Check the platform name to determine the Embeddings class to set in Definition
• Change the name config entry to className and allow users to pass PhpLlm\LlmChain\Platform\Bridge\Mistral\Embeddings as value. In LlmChainExtension we could check the class existence and if it extends PhpLlm\LlmChain\Platform\Model.

@chr-hertel @OskarStark you probably have a clear idea of what is the best solution here, maybe another one?

---

Backported in symfony/ai#59