C-Pack: Packaged Resources To Advance General Chinese Embedding
Paper • 2309.07597 • Published • 2
bge-small-en-v1.5 for sidekick (Core ML)
BAAI/bge-small-en-v1.5, a BERT embedding model (384 dimensions), converted to Core ML by sidekick for Apple silicon's Neural Engine, GPU and CPU. This repository holds one build, fp16/, graded against the original model on every compute path it runs on.
Licence: MIT · publishing tier 2a. The card permits commercial use, but neither it nor the C-Pack paper names the English pre-training corpus or the sources of the ~200M English training pairs. Undisclosed training data: 2a. Whether a dataset's terms reach a model trained on it is legally unsettled; the training data below gives each source's licence. How sidekick decides what it publishes.
Download
| Build | Weights | Sequence buckets | Max tokens | Download | Shared weights | Runs on (grade) |
|---|---|---|---|---|---|---|
fp16/ |
16-bit float | 128, 256, 512 | 512 | 68.5 MB | Yes, one file for all buckets | Neural Engine (default) A, GPU A, CPU A |
sidekickd fetch bge-small-en-v1.5
Needs sidekick 0.16.0 or later. How fetch works · Installing without fetch · Calling the model · Why compiled programs
Fidelity
Graded against the original model in fp32 by sidekick's parity suite. What the grades mean.
| Compute path | Chip | OS | Grade | Worst cosine | Graded files |
|---|---|---|---|---|---|
| Neural Engine (default) | Apple M1 Max | macOS 27.0 | A | 0.99997 | This build |
| GPU | Apple M1 Max | macOS 27.0 | A | 0.9999991 | This build |
| CPU | Apple M1 Max | macOS 27.0 | A | 0.99993 | This build |
| GPU | Intel Xeon W-3235 | macOS 26.6.2 | A | 0.999999 | Unconfirmed |
| CPU | Intel Xeon W-3235 | macOS 26.6.2 | A | 0.999998 | Unconfirmed |
This build: the grading report's file hashes match this repository's files. Unconfirmed: graded on a build of the same model that can't be matched to these files by hash.
Benchmark scores
The original model's published scores, not this build's. Every score, with its caveats.
| Benchmark | Score | Measure | Source |
|---|---|---|---|
| MTEB(eng, v2) | 64.30 | Mean over its tasks | MTEB results, some tasks self-reported |
| MTEB(Multilingual, v2) | 43.76 | Mean over its tasks | MTEB results |
| MTEB(eng, v1) | 61.75 | Mean over its tasks | MTEB results, some tasks self-reported |
Training data
Training data: undisclosed. Each dataset under its original source's licence:
| Dataset | Licence | Terms | Stage |
|---|---|---|---|
| BAAI RetroMAE pre-training corpus (English) | unknown | Unknown | pre-training (RetroMAE) |
| BAAI English contrastive pairs (~200M pairs, released 'together with C-MTP') | unknown | Unknown | contrastive fine-tune |
| BAAI English labelled fine-tuning data | unknown | Unknown | supervised fine-tune |
Not confirmed from a primary source
- Contents and licences of the BAAI-MTP English data (data.baai.ac.cn page did not render).
- Whether bge-small-en-v1.5 was initialised from a public BERT checkpoint (not stated on the card).
- C-Pack paper content read through a summarising fetch of its arXiv HTML.
Original model
| Model | BAAI/bge-small-en-v1.5 |
| Revision | 5c38ec7c405ec4b44b94cc5a9bb96e735b38267a |
| Licence | MIT |
| Licence file | LICENSE: the upstream licence text, with its copyright line, from FlagOpen/FlagEmbedding |
| Base model | None declared |
| Changes in this repository | Converted to Core ML in 16-bit float by tools/convert_bge_small.py, with a sidekick manifest added. Reproduce it |
Checksums
fp16/: ledger build bge-small-en-v1.5/coreml/fp16/f1b85f22e0b4, 17 files, 1 shared
sidekickd fetch checks every file against sidekick's checksum ledger; fp16.sha256 has the same list for shasum -a 256 -c.
| Built with | |
|---|---|
| Converter | tools/convert_bge_small.py at 761deb3f2e23 |
| coremltools | 9.0 |
| PyTorch | 2.13.0 |
| Python | 3.12.13 |
| OS | macOS 27.0 (26A428) |
| Xcode | 27.1 (27A9275) |
| Converted on | Apple M1 Max |
| Converted | 2026-10-23 |
| Shared weights | Yes: one weights file for every bucket |
| sha256 | Size | Path |
|---|---|---|
3b9e5783ff20962eb45d0784f565352bf9b77370c1032d2f8a00f7a05a2234c6 |
2.8 kB | manifest.toml |
1cb304b8dfd1d71dafa83762bfd53755f2f4bb4df6277a0b58871e2ccd8a030c |
243 B | model_128.mlmodelc/analytics/coremldata.bin |
16cb073a96ae4ab640b6b38a7c30429259748a983469920d13bca19dbbf55027 |
409 B | model_128.mlmodelc/coremldata.bin |
c23c956c5dace7c3d47d2e792ef7c65f23623c493ffe4e0f9ec3d579b37863f0 |
2.3 kB | model_128.mlmodelc/metadata.json |
7ae733bc38b2f663ec263bcd7bc0ef0fb7209948961bf28c871f6733157e2fe8 |
130.2 kB | model_128.mlmodelc/model.mil |
67cca0adb2abb0b47925800e1f6914261fac21f3dcf92412a4a81cd8041cae44 |
67.4 MB | model_128.mlmodelc/weights/weight.binmodel_256.mlmodelc/weights/weight.binmodel_512.mlmodelc/weights/weight.bin |
2e13be73b30a5d8c4e99a84a865a5c1c617eafccb883b7f03abe062f08ff3400 |
243 B | model_256.mlmodelc/analytics/coremldata.bin |
d90ce7f35844742028c9614ab639ce1550c1e803381850016bafce71fd1f2708 |
409 B | model_256.mlmodelc/coremldata.bin |
adc812afd02b558ee99954577a4dbb263788d77b7fb7973e8d2ffa79436be325 |
2.3 kB | model_256.mlmodelc/metadata.json |
4839bc4064e31bb639e5bc2af9c25937621410df276a5b75cb0dd607a2eb6b58 |
130.2 kB | model_256.mlmodelc/model.mil |
b649e3e4e8c2d648490f73719c8648ad167e21d54f7c03ac6aeacf4600b18df3 |
243 B | model_512.mlmodelc/analytics/coremldata.bin |
66da60a707ee4fd1b9e861f61fb74a7af41ee2534e3ff037c930761bb833688a |
409 B | model_512.mlmodelc/coremldata.bin |
5bdf674443f109e1ce2515305fef0194832e224d74b943f00a91c2dc6893592b |
2.3 kB | model_512.mlmodelc/metadata.json |
8d6964cc402c30986ccc85a86d06d1e12b106ef9fcc7389527f80b59b3a6f4fe |
130.2 kB | model_512.mlmodelc/model.mil |
d241a60d5e8f04cc1b2b3e9ef7a4921b27bf526d9f6050ab90f9267a1f9e5c66 |
711.4 kB | tokenizer.json |
Generated by tools/model_card.py from sidekick's committed data. Report a problem at https://github.com/ironblock/sidekick/issues.
Model tree for sidekickd/bge-small-en-v1.5-coreml
Base model
BAAI/bge-small-en-v1.5