limbed
ColBERT embedding and MaxSim scoring without PyTorch. Uses a native C++ extension (ONNX Runtime + tokenizers-cpp) so you don't need to pull in 2 GB of deep learning dependencies just to encode some text.
Install
pip install limbed-py
Only runtime deps are numpy and huggingface-hub.
Usage
from limbed import LateEmbedder, compute_maxsim
model = LateEmbedder() # downloads thlurte/mxbai-edge-colbert-v0-17m-onnx
q = model.encode_queries("What is late interaction?")
d = model.encode_docs("ColBERT computes token-level similarity.")
score = compute_maxsim(q[0], d[0])
print(score)
You can also point it at a local directory with model.onnx and tokenizer.json:
model = LateEmbedder("./my-model/")
Late chunking (long documents)
For RAG, encode the full document (in model-sized windows when needed), then slice retrieval chunks from the contextualized token embeddings. Requires pip install tokenizers.
from limbed import LateEmbedder, compute_maxsim_late_chunked
model = LateEmbedder("jina-colbert-v2") # 8192 context; BERT models use 512
doc = "..." # full document
chunks = ["First paragraph.", "Second paragraph."] # substrings of doc
chunk_embs = model.encode_doc_late_chunked(doc, chunks)
q = model.encode_queries("your query")
score = compute_maxsim_late_chunked(q[0], chunk_embs) # max over chunks
Documents longer than the model limit are encoded in overlapping windows (window_overlap=64 by default). Chunks entirely inside one window retain full within-window context.
Models
| Alias | Repo | Size | Dim | Notes |
|---|---|---|---|---|
mxbai-edge-colbert-v0-17m |
thlurte/mxbai-edge-colbert-v0-17m-onnx |
66 MB | 48 | Default |
mxbai-edge-colbert-v0-32m |
thlurte/mxbai-edge-colbert-v0-32m-onnx |
124 MB | 64 | |
colbertv2.0 |
thlurte/colbertv2.0-onnx |
438 MB | 128 | Standard ColBERTv2.0 BERT-based model |
answerai-colbert-small-v1 |
thlurte/answerai-colbert-small-v1-onnx |
135 MB | 96 | Lightweight, high-performance model |
jina-colbert-v2 |
thlurte/jina-colbert-v2-onnx |
2.23 GB | 128 | XLM-RoBERTa multilingual model |
lateon |
thlurte/lateon-onnx |
580 MB | 128 | Case-sensitive: use do_lower_case=False |
Any ColBERT ONNX model should work if you put model.onnx and tokenizer.json in a folder and pass the path.
Benchmarks
The following benchmark was run on CPU using 20 queries (max length 32) and 20 documents (max length 256), comparing limbed against fastembed execution:
Performance (Throughput & Speedup)
| Model | Operation | limbed Throughput |
fastembed Throughput |
Speedup (Wall-clock) |
|---|---|---|---|---|
| ColBERTv2.0 | Queries | 71.3 QPS | 31.6 QPS | 2.25x |
| ColBERTv2.0 | Documents | 93.8 DPS | 66.1 DPS | 1.42x |
| Jina ColBERT v2 | Queries | 6.2 QPS | 5.0 QPS | 1.25x |
| Jina ColBERT v2 | Documents | 10.1 DPS | 5.2 DPS | 1.94x |
How it works
- Tokenization and inference run in C++ via a nanobind extension
- GIL is released during encode and MaxSim calls, so you can run multiple threads
- Punctuation tokens are masked out of document embeddings (standard ColBERT behavior)
- Embeddings are L2-normalized by default
- CPU only for now
Docker & Alpine Linux Compatibility
Because the underlying precompiled ONNX Runtime library is linked against glibc, this package will not run out-of-the-box on Alpine Linux images (e.g., python:3.10-alpine).
If deploying via Docker, it is highly recommended to use a Debian-based slim image:
FROM python:3.10-slim
If you must use Alpine, you will need to install the compatibility layer: apk add --no-cache gcompat.
Supported Platforms & Architectures
Precompiled wheels are published to PyPI for the following environments:
| Operating System | Architecture | Python Versions | Notes |
|---|---|---|---|
| Linux | x86_64, aarch64 |
3.10, 3.11, 3.12, 3.13, 3.14 | Built on manylinux_2_28 (glibc-based) |
| macOS | arm64 (Apple Silicon) |
3.10, 3.11, 3.12, 3.13, 3.14 | SDK/deployment target macOS 13.3+ |
| Windows | AMD64 (x86_64) |
3.10, 3.11, 3.12, 3.13, 3.14 |
[!NOTE] Other platforms (such as Intel-based macOS or ARM-based Windows) will fall back to compilation from the source distribution (
sdist). This requires a local C++ compiler (supporting C++17) and CMake.
License
MIT. See LICENSE.
1 maintainer
Release files for limbed-py 0.1.7
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| limbed_py-0.1.7.tar.gz | 25.7 kB | Details |
Built distributions (wheels)
Total release size: 253.7 MB
Release files / limbed_py-0.1.7.tar.gz
| Download URL | limbed_py-0.1.7.tar.gz |
|---|---|
| Size | 25.7 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
75d1ee9d738084de306493e1e2fc299155f4c37d4c898904a819bc33b05d5a3e
|
|
BLAKE2b-256 checksum How to use checksums |
8df2e9235e53d88eef4905eb27ab2eec7140f7d95b6537edbc9bc655839253ad
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 28, 2026.
Transparency logRelease files / limbed_py-0.1.7-cp313-cp313-win_amd64.whl
| Download URL | limbed_py-0.1.7-cp313-cp313-win_amd64.whl |
|---|---|
| Size | 6.7 MB |
| Tags | CPython 3.13 Windows x86-64 |
|
SHA-256 checksum How to use checksums |
f5ec7f4bbfde84a029d05f27c02da1cf4d1dc4143bff996002014eca774fc1ae
|
|
BLAKE2b-256 checksum How to use checksums |
35de695508880b8956c209149e290b9b719ed4f3160a5180248f94abb2cf0d77
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 28, 2026.
Transparency logRelease files / limbed_py-0.1.7-cp313-cp313-manylinux_2_28_x86_64.whl
| Download URL | limbed_py-0.1.7-cp313-cp313-manylinux_2_28_x86_64.whl |
|---|---|
| Size | 21.4 MB |
| Tags | CPython 3.13 Linux glibc 2.28+ x86-64 |
|
SHA-256 checksum How to use checksums |
1554d7192f7cd43cec72eda5013454a31b330015d100bb10d6b35326e4ec3682
|
|
BLAKE2b-256 checksum How to use checksums |
fc815f7d469297b9dc5ef0424ca6f7be3fe2f6de690a3239325f613d39f68a7c
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 28, 2026.
Transparency logRelease files / limbed_py-0.1.7-cp313-cp313-manylinux_2_28_aarch64.whl
| Download URL | limbed_py-0.1.7-cp313-cp313-manylinux_2_28_aarch64.whl |
|---|---|
| Size | 18.8 MB |
| Tags | CPython 3.13 Linux glibc 2.28+ ARM64 |
|
SHA-256 checksum How to use checksums |
4cf9cc795d1a2c0e5d9b7274c2f72597f35d19f837661a1ad6925b9cfd9197f7
|
|
BLAKE2b-256 checksum How to use checksums |
db698232d880bdbae39960e976888b94205bb4316ff25ff50f952cca780697b3
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 28, 2026.
Transparency logRelease files / limbed_py-0.1.7-cp313-cp313-macosx_13_0_arm64.whl
| Download URL | limbed_py-0.1.7-cp313-cp313-macosx_13_0_arm64.whl |
|---|---|
| Size | 16.6 MB |
| Tags | CPython 3.13 macOS 13.0+ ARM64 |
|
SHA-256 checksum How to use checksums |
01c30d63f4852dd09603a2813b685998353562583b25792d96d81590ebc798a4
|
|
BLAKE2b-256 checksum How to use checksums |
bc02831981e4c658f4e741b107fa2dbaaf0d4a988047f61a024c96c349298633
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 28, 2026.
Transparency logRelease files / limbed_py-0.1.7-cp312-cp312-win_amd64.whl
| Download URL | limbed_py-0.1.7-cp312-cp312-win_amd64.whl |
|---|---|
| Size | 6.7 MB |
| Tags | CPython 3.12 Windows x86-64 |
|
SHA-256 checksum How to use checksums |
e51583e573d872cd6179449610dcfbb24021e92651f5c509e44eae54af6f1ad7
|
|
BLAKE2b-256 checksum How to use checksums |
2a867700e4bb4a0ee6f6179d766c3e4cca37d5824d4680e9d23118abbb71b86e
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 28, 2026.
Transparency logRelease files / limbed_py-0.1.7-cp312-cp312-manylinux_2_28_x86_64.whl
| Download URL | limbed_py-0.1.7-cp312-cp312-manylinux_2_28_x86_64.whl |
|---|---|
| Size | 21.4 MB |
| Tags | CPython 3.12 Linux glibc 2.28+ x86-64 |
|
SHA-256 checksum How to use checksums |
8412c66660974cde192b7e704fbf22221af2f1a7ad733e876f39eb82668000fd
|
|
BLAKE2b-256 checksum How to use checksums |
036081a7fb9f8b63efabec8da0f1fba0ca7a3b3cba11c58683bcf5ebd07ce76c
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 28, 2026.
Transparency logRelease files / limbed_py-0.1.7-cp312-cp312-manylinux_2_28_aarch64.whl
| Download URL | limbed_py-0.1.7-cp312-cp312-manylinux_2_28_aarch64.whl |
|---|---|
| Size | 18.8 MB |
| Tags | CPython 3.12 Linux glibc 2.28+ ARM64 |
|
SHA-256 checksum How to use checksums |
751f70ebac88c7b0395d3e7ecc84c4982e3e0cbd4e11e548c28da56614fc1add
|
|
BLAKE2b-256 checksum How to use checksums |
f6647627e61ba8e41ae1ea2e80c0006d5e79e24080aaad60cb9f6b1f1e8927c8
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 28, 2026.
Transparency logRelease files / limbed_py-0.1.7-cp312-cp312-macosx_13_0_arm64.whl
| Download URL | limbed_py-0.1.7-cp312-cp312-macosx_13_0_arm64.whl |
|---|---|
| Size | 16.6 MB |
| Tags | CPython 3.12 macOS 13.0+ ARM64 |
|
SHA-256 checksum How to use checksums |
becb5dfdf3ba97e6d5a270ae22efb7b5f0f3d10d2c6ba587794f918f1dcc2b5e
|
|
BLAKE2b-256 checksum How to use checksums |
c79137cba17d8a2272f614127766946376bd997eb9c3a6a9aa9ef5f61f5392b1
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 28, 2026.
Transparency logRelease files / limbed_py-0.1.7-cp311-cp311-win_amd64.whl
| Download URL | limbed_py-0.1.7-cp311-cp311-win_amd64.whl |
|---|---|
| Size | 6.7 MB |
| Tags | CPython 3.11 Windows x86-64 |
|
SHA-256 checksum How to use checksums |
25c915a758349a8fdc61b18f44431b4ec9f7b5c0e9de30807d9e8af1a262ccc8
|
|
BLAKE2b-256 checksum How to use checksums |
bfa693cc0bf4b892a539a4567f1857f3fb728e4ef78e3075b36c7f9560010a77
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 28, 2026.
Transparency logRelease files / limbed_py-0.1.7-cp311-cp311-manylinux_2_28_x86_64.whl
| Download URL | limbed_py-0.1.7-cp311-cp311-manylinux_2_28_x86_64.whl |
|---|---|
| Size | 21.4 MB |
| Tags | CPython 3.11 Linux glibc 2.28+ x86-64 |
|
SHA-256 checksum How to use checksums |
973c3986adb44c6b5347b86c3d25d4c883df68aaf4f6f9bec39a01bdbec3f7e7
|
|
BLAKE2b-256 checksum How to use checksums |
1b55dc58198a62223c6a53e5650ed2aeee3ad751f9fe0e6ad3cff9b6cb754e8d
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 28, 2026.
Transparency logRelease files / limbed_py-0.1.7-cp311-cp311-manylinux_2_28_aarch64.whl
| Download URL | limbed_py-0.1.7-cp311-cp311-manylinux_2_28_aarch64.whl |
|---|---|
| Size | 18.8 MB |
| Tags | CPython 3.11 Linux glibc 2.28+ ARM64 |
|
SHA-256 checksum How to use checksums |
bab65a72e0a64adf01ac89d268c3f027839bd824c573b3d1a60517c3894142aa
|
|
BLAKE2b-256 checksum How to use checksums |
227e989763b9f600a01f4793006c824678fa68b20c8c782058cd2064c60705b4
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 28, 2026.
Transparency logRelease files / limbed_py-0.1.7-cp311-cp311-macosx_13_0_arm64.whl
| Download URL | limbed_py-0.1.7-cp311-cp311-macosx_13_0_arm64.whl |
|---|---|
| Size | 16.6 MB |
| Tags | CPython 3.11 macOS 13.0+ ARM64 |
|
SHA-256 checksum How to use checksums |
2d5db5f181a8dd0bb99015114d5a435f30153863d1c9cf7a75e91061104f3049
|
|
BLAKE2b-256 checksum How to use checksums |
308b735b19487034eee4288470fbb35dd8fde3c2659a32742fc1954ae0ad628c
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 28, 2026.
Transparency logRelease files / limbed_py-0.1.7-cp310-cp310-win_amd64.whl
| Download URL | limbed_py-0.1.7-cp310-cp310-win_amd64.whl |
|---|---|
| Size | 6.7 MB |
| Tags | CPython 3.10 Windows x86-64 |
|
SHA-256 checksum How to use checksums |
2493f68c05269577dfb12eb33fbf57e0d5e94f57f1bf25afb94d3009edea9a00
|
|
BLAKE2b-256 checksum How to use checksums |
e7d9610d80edc100c1c65c514afbbfa13ee0a774c21d2f3b31550ada3736fbc1
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 28, 2026.
Transparency logRelease files / limbed_py-0.1.7-cp310-cp310-manylinux_2_28_x86_64.whl
| Download URL | limbed_py-0.1.7-cp310-cp310-manylinux_2_28_x86_64.whl |
|---|---|
| Size | 21.4 MB |
| Tags | CPython 3.10 Linux glibc 2.28+ x86-64 |
|
SHA-256 checksum How to use checksums |
55c53d3728540664c972585f65bb9fd19fa795c968a8ac5a9e9c784b9f1718f6
|
|
BLAKE2b-256 checksum How to use checksums |
3c87b3787a40af8032f6f76b743253e830da1df7464da981e7f1433c19db2e6c
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 28, 2026.
Transparency logRelease files / limbed_py-0.1.7-cp310-cp310-manylinux_2_28_aarch64.whl
| Download URL | limbed_py-0.1.7-cp310-cp310-manylinux_2_28_aarch64.whl |
|---|---|
| Size | 18.8 MB |
| Tags | CPython 3.10 Linux glibc 2.28+ ARM64 |
|
SHA-256 checksum How to use checksums |
87df9091dea89d66e82e14c7f1d2b90f07ddf0429ca4920b040a77646e082954
|
|
BLAKE2b-256 checksum How to use checksums |
392269e9c26fe63b4ab7b92ac0f91d2d9bd38bc9132b43d0a74655945611290b
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 28, 2026.
Transparency logRelease files / limbed_py-0.1.7-cp310-cp310-macosx_13_0_arm64.whl
| Download URL | limbed_py-0.1.7-cp310-cp310-macosx_13_0_arm64.whl |
|---|---|
| Size | 16.6 MB |
| Tags | CPython 3.10 macOS 13.0+ ARM64 |
|
SHA-256 checksum How to use checksums |
d8af630df9637d34b246e70ba2213f20be7b9a4e4f499831d81b2ade73e6ec3c
|
|
BLAKE2b-256 checksum How to use checksums |
f756f098a2625854729bccad1e2ce4d70d1955412f423d7ba76655525b01fb89
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 28, 2026.
Transparency log
