Is it possible to serve EmbeddingGemma in /embeddings using mlx-serve?

[ajram23]

Looks like it still needs

Parse and honor:

• use_bidirectional_attention: true
• tokenizer add_bos_token / add_eos_token or tokenizer post-processor

Load/apply EmbeddingGemma projection:

• dense.0
• dense.1
• then existing mean-pool + L2-normalize

[ddalcu]

yep, should be pretty easy, do you want to attempt it and open a PR :) ?