feat: add Jina AI embedding provider

[hanxiao]

Add Jina AI as a supported embedding provider for dense embeddings.

MMTEB scores vs model size. jina-v5-text models (red) outperform models 2-16x their size.

MTEB English v2 scores. v5-text-nano (239M) achieves 71.0, matching models with 2x+ parameters.

Both models are open-weight (Apache 2.0) and support Matryoshka dimension reduction, task-specific embeddings, and local deployment via GGUF/MLX.

Paper: arXiv:2602.15547 | Blog | HuggingFace

Features

• API mode: Jina AI hosted API at https://api.jina.ai/v1 (OpenAI-compatible)
• Local mode: Open-weight models available in GGUF and MLX formats on HuggingFace. Run locally with llama.cpp, MLX, or vLLM and point api_base to your local server.
• Task-specific embeddings via task parameter
• Late chunking support via late_chunking parameter
• Matryoshka dimension reduction via dimensions parameter

Changes

• New jina_embedders.py with JinaDenseEmbedder
• Register jina provider in embedding_config.py
• Update docs (EN/ZH), README, FAQ with Jina provider info
• Add unit tests with mocked API calls

[CLAassistant]

All committers have signed the CLA.

[hanxiao]

@MaojiaSheng @ZaynJarvis Could you review this PR when you get a chance? Thanks!

[ZaynJarvis]

looks like jina cannot set other dimension except the configured ones, consider remove this config? (includes docs) also does JinaEmbedder has dimension value validation?

Other lgtm

[hanxiao]

v5 supports arbitrary dimensions via MRL - any value (1, 33, 34, 512, etc.) up to model max works. max is per-model: 1024 for small, 768 for nano. validation added.

[hanxiao]

Thanks for the review! Regarding the dimension config:

Jina v5 models support Matryoshka Representation Learning (MRL), so the dimensions parameter can truncate embeddings to any value up to the max dimension (1024 for small, 768 for nano). The API handles truncation + L2 renormalization server-side.

I will add dimension validation in JinaDenseEmbedder.__init__ to ensure the requested dimension does not exceed the model max. Pushing the fix now.

[ZaynJarvis]

Thx, lgtm @MaojiaSheng

[MaojiaSheng]

Thanks, will be merged