kevinqz/BGE-Small-EN-v1.5-CoreAI

Canonical: kevinqz/BGE-Small-EN-v1.5-CoreAI — source of truth.

BGE-Small-EN v1.5 (fabric)

An Apple Core AI conversion of BAAI/bge-small-en-v1.5 — a text-embedding encoder that maps (input_ids, attention_mask) to a pooled, L2-normalized sentence embedding (cosine-ready). Produced by coreai-fabric and indexed by coreai-catalog.

Encoder, not a chat model. This is a single-forward encoder — token ids in, one 384-d L2-normalized embedding (1×384) out. No text generation, no KV-cache. The host owns the upstream tokenizer and pad/truncate to the static sequence length; compare embeddings with a dot product (cosine).

Model facts

Field	Value
Parameters	0.033B
Architecture	encoder
Capabilities	embedding
Embedding dim	384
Sequence length	128 (static)
Pooling	cls, L2-normalized
Quantization / precision	none / float32
On-disk size	126 MB
Asset kind	single-graph encoder ((input_ids, attention_mask) -> unit embedding)
assetVersion	2.0

Use it — this needs host code you supply

The bundle is a single static-sequence graph: (input_ids, attention_mask) [1,128] in → embedding 1×384 out (cls pooling, unit-norm). You supply the upstream tokenizer and pad/truncate in your host code (Swift or Python). Token ids are int32 at the graph boundary.

pip install coreai-catalog && coreai-catalog install bge-small-en-v15

Requirements

• Deployment: macOS 27.0+ / iOS 27.0+, Xcode 27+. The asset serializes with minimum_os v27, so the on-device Swift runtime requires macOS/iOS 27+. A Mac on macOS 26 can convert and inspect it but not run it on-device.
• Apple Silicon.

Verification (output parity)

• Gate A (structure): passed — the bundle's layout + metadata were validated; the graph loads.
• Gate B — graph_output_cosine: 1.000000 min output cosine (median 1.000000) vs the fp32 torch sentence encoder over 8 seeded (input_ids, attention_mask), measured on apple_silicon. Certifies the export computes the SAME output as the source — a conversion-fidelity metric, not task accuracy.
• This certifies the export is numerically faithful to the source encoder — it does NOT certify retrieval quality on your corpus. Reproduce with coreai-fabric verify.

Provenance

Field	Value
Base model	BAAI/bge-small-en-v1.5 @ 5c38ec7c405ec4b44b94cc5a9bb96e735b38267a
Converted by	models/embedding/export.py (version not reported)
Recipe	bge-small-en-v15 (recipe_source: fabric)
Precision / quantization	float32 / none
Conversion date	2026-07-10

Machine-readable, in this repo: parity-report.json · reproduce-manifest.json · LICENSE.

License and attribution

Weights licensed mit — see the bundled LICENSE. This artifact is a converted derivative of the base model: its weights were converted to Apple Core AI format. The conversion itself is community work.

Links

• Base model: BAAI/bge-small-en-v1.5
• Reproduce: recipe bge-small-en-v15
• Index: coreai-catalog
• HF Collection

The on-device Core AI ecosystem

• coreai-fabric — the reproducible recipe → .aimodel pipeline that produced this asset.
• coreai-catalog — the index of Core AI models with provenance and integration snippets.
• apple/coreai-models — Apple's official exporters and runtimes.

Not affiliated with Apple

Community conversion. Not produced, hosted, or endorsed by Apple. Apple and Core AI are trademarks of Apple Inc., used here only to describe the target runtime/format.