jinaai/jina-embeddings-v5-text-nano-retrieval-mlx

fix prefix: add Document: prefix for non-retrieval tasks

revert model.py to nano-specific implementation

fix: handle models without QK norm (EuroBERT/nano)

optimize: use mx.fast.rope and mx.fast.scaled_dot_product_attention

update model.py: fix rope_theta comment (1000000)

reconvert weights: PyTorch FP32 -> MLX FP16, fix rope_theta=1000000

Fix rope_theta default: 250000 -> 1000000

Convert weights to float32 for better precision (was float16)

fix rope_theta: 250000 -> 1000000

fix README rendering: pure markdown, no HTML details tags

rewrite README to match upstream format

fix usage: git clone + local paths, no huggingface_hub dependency

fix usage: add huggingface_hub download, show full runnable example

fix README: match v5-small-mlx style, correct bibtex and usage

Initial upload: MLX float16

initial commit