Model card
Browse files
README.md
ADDED
|
@@ -0,0 +1,49 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-nc-4.0
|
| 3 |
+
library_name: coreml
|
| 4 |
+
tags:
|
| 5 |
+
- coreml
|
| 6 |
+
- embeddings
|
| 7 |
+
- feature-extraction
|
| 8 |
+
- jina-embeddings-v5
|
| 9 |
+
- apple-silicon
|
| 10 |
+
- ane
|
| 11 |
+
---
|
| 12 |
+
|
| 13 |
+
# jina-code-embeddings-0.5b — Core ML bundle
|
| 14 |
+
|
| 15 |
+
On-device Core ML conversion of **[jinaai/jina-code-embeddings-0.5b](https://huggingface.co/jinaai/jina-code-embeddings-0.5b)**,
|
| 16 |
+
produced by the [Glossematics](https://huggingface.co/1of2) conversion pipeline. The repository
|
| 17 |
+
root is a valid **model bundle** (`manifest.json`, formatVersion 2) consumed directly by the
|
| 18 |
+
Glossematics Swift package — download the snapshot and point `GlossModelBundle` at it.
|
| 19 |
+
|
| 20 |
+
- Embedding dimension: **896** (Matryoshka: [64, 128, 256, 512, 896])
|
| 21 |
+
- Towers: text
|
| 22 |
+
- Text buckets: [32, 64, 128, 256, 512, 1024, 2048] + batch-4 functions at [32, 64, 128]
|
| 23 |
+
- Bidirectional text tower with explicit attention mask: False
|
| 24 |
+
- Artifacts: precompiled .mlmodelc (loads directly, no on-device compile)
|
| 25 |
+
- Numeric recipe: **float16 weights / float16 activations** — embedding space `jinaai/jina-code-embeddings-0.5b:896:w16a16`
|
| 26 |
+
|
| 27 |
+
## Usage (Swift, macOS 15 / iOS 18)
|
| 28 |
+
|
| 29 |
+
```swift
|
| 30 |
+
import Glossematics
|
| 31 |
+
|
| 32 |
+
let bundle = try await GlossModelBundle(hub: "1of2/jina-code-embeddings-0.5b-coreml-w16")
|
| 33 |
+
let gloss = GlossOmniEmbedder(bundle: bundle)
|
| 34 |
+
let q = try await gloss.embed(text: "What is the capital of France?", prompt: .query)
|
| 35 |
+
```
|
| 36 |
+
|
| 37 |
+
## License
|
| 38 |
+
|
| 39 |
+
The source model weights are **CC-BY-NC-4.0** (© Jina AI) and this converted artifact is a
|
| 40 |
+
derivative work distributed under the same license: **non-commercial use only**, with
|
| 41 |
+
attribution. Commercial use requires a license from Jina AI. See the
|
| 42 |
+
[source model card](https://huggingface.co/jinaai/jina-code-embeddings-0.5b) for details.
|
| 43 |
+
|
| 44 |
+
## Provenance & validation
|
| 45 |
+
|
| 46 |
+
- Source: `jinaai/jina-code-embeddings-0.5b` revision `4db235132dafbe56a8b9c5f59b59795ecf58a4a7`
|
| 47 |
+
- Converter: `jina_convert` 0.2.0 (commit 4fd131b)
|
| 48 |
+
- Validated against fp32 references captured from the source model (per-modality cosine parity
|
| 49 |
+
gates; end-to-end through the public Swift API via `gloss-bundle-smoke`).
|